Integrating Machine Learning in Environmental DNA Metabarcoding for Improved Biodiversity Assessment: A Review and Analysis of Recent Studies

doi:10.21203/rs.3.rs-2823060/v1

Download PDF

Research Article

Integrating Machine Learning in Environmental DNA Metabarcoding for Improved Biodiversity Assessment: A Review and Analysis of Recent Studies

https://doi.org/10.21203/rs.3.rs-2823060/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

A subset of artificial intelligence (AI) known as "machine learning" (ML) allows computer systems to learn from experience and advance without explicit programming. Environmental DNA (eDNA) metabarcoding is a rapidly expanding scientific method for determining species' existence in the environment. It includes sequencing DNA from environmental samples. This work investigates how machine learning (ML) could be used in eDNA metabarcoding to enhance biodiversity estimates.

Reviewing current work on the use of ML in eDNA metabarcoding, this paper focuses on the various ML algorithms utilized, the varied kinds of data inputs, and the advances in biodiversity assessment that occur. In 23 investigations, machine learning (ML) was used for Environmental DNA (eDNA) metabarcoding.

In terms of increasing the precision and effectiveness of eDNA metabarcoding for biodiversity assessments, ML has shown encouraging outcomes. In addition to accurately classifying and predicting species from eDNA sequences, ML algorithms may uncover uncommon or invasive species often overlooked by conventional techniques. According to the research we looked at, compared to conventional approaches, there was an average increase of 20% in detection sensitivity and 14% in species richness.

Adding ML to eDNA metabarcoding has enormous promise for enhancing ecological monitoring and biodiversity assessments. Nonetheless, it is essential to choose suitable ML algorithms, ensure there are enough training datasets, and assess the accuracy of the predictions made by ML. This research underscores the need for more investigation to fully fulfill ML's potential for ecological monitoring and conservation while providing a thorough review of its possible uses in eDNA metabarcoding.

machine learning

environmental DNA

metabarcoding

biodiversity assessment

artificial intelligence

ecological monitoring

species detection

data analysis

ML algorithms

conservation

Direct species sampling is a step in the traditional techniques of evaluating biodiversity that may be labor- and time-intensive and provide findings that need to be completed or skewed. In recent years, environmental DNA (eDNA) metabarcoding has become a powerful and non-intrusive molecular method for biodiversity assessment. This method uses DNA sequencing to determine which species are present in the investigated ecosystem from environmental samples like soil, water, or air (Taberlet et al., 2018).

eDNA metabarcoding has numerous drawbacks despite its promise. The quality and amount of DNA samples, the selection of genetic markers, and the data processing techniques may all impact the accuracy and efficacy of eDNA metabarcoding for species detection and identification (Deiner et al., 2017). In order to enhance biodiversity assessments, machine learning (ML) techniques have been investigated for integration with eDNA metabarcoding.

A subtype of artificial intelligence called "machine learning" enables computer systems to learn from their past performance and advance without explicit programming. ML algorithms are potentially valuable tools for eDNA metabarcoding since they can be trained on big datasets to find patterns, categorize data, and predict results. ML algorithms have been used to find uncommon or invasive species often overlooked by conventional techniques, categorize and predict species using eDNA sequences, and identify them (Elbrecht & Leese, 2017; Ji et al., 2021).

In this review, we want to thoroughly summarize current research that has employed ML algorithms in eDNA metabarcoding to quantify biodiversity more accurately. We will discuss the various ML algorithms used, the kinds of data inputs, and the improvements in biodiversity assessment. We will also talk about the gaps and difficulties in this area and some future possibilities for ML-based eDNA metabarcoding.

In recent years, eDNA metabarcoding has drawn much interest as a quick and effective way to measure biodiversity. Direct species sampling is used in traditional techniques of measuring biodiversity, which may be labor- and time-intensive and provide findings that need to be completed or skewed (Taberlet et al., 2018). In contrast, eDNA metabarcoding uses environmental materials like soil, water, or air to determine which species are present in the tested ecosystem. The method uses genetic markers such as mitochondrial DNA or ribosomal RNA to identify the species present by sequencing the DNA from these samples and comparing the sequences to reference databases (Deiner et al., 2017).

Although eDNA metabarcoding has several benefits over conventional techniques, there are also some drawbacks. These restrictions include the likelihood of amplifying non-target DNA, the probability of false positives or negatives due to contamination, eDNA degradation, or low abundance in samples (Deiner et al., 2017). Moreover, eDNA metabarcoding data processing techniques may be intricate and need careful consideration of several variables, including sequence quality, reference databases, and statistical analysis (Elbrecht & Leese, 2017).

In order to enhance biodiversity assessments, there has been an increase in interest in ML algorithms' integration with eDNA metabarcoding in recent years. By allowing more precise and effective species recognition and identification, machine learning methods have the potential to get around some of the drawbacks of eDNA metabarcoding. ML use in eDNA metabarcoding must overcome some obstacles to fully achieve its potential since it is still in its early phases.

This review aims to provide a thorough overview of current research that has employed ML algorithms in eDNA metabarcoding for more accurate biodiversity assessment. We specifically want to identify the various ML algorithms utilized, the kinds of data inputs, and the advancements in biodiversity assessment that come from them. We will discuss the difficulties and limitations in this study area and provide suggestions for further investigation.

The evaluation of biodiversity might be changed entirely by using ML algorithms in eDNA metabarcoding. Large datasets may be used to train ML algorithms to discover uncommon or invasive species that are often overlooked by conventional approaches and categorize and predict species using eDNA sequences. By increasing the effectiveness of data processing, ML algorithms may also cut down on the time and labor needed for biodiversity assessments. That has significant ramifications for ecological monitoring and conservation initiatives, making conducting more precise and effective biodiversity evaluations across various environments possible.

The development of more precise and effective machine learning algorithms, the need for richer training datasets, and the need for standardized methods for ML-based eDNA metabarcoding are some of the current gaps and problems in this sector. These difficulties must be overcome to fully fulfill the promise of ML-based eDNA metabarcoding for biodiversity assessment.

This paper emphasizes the potential for eDNA metabarcoding to use ML algorithms for enhanced biodiversity assessment. In terms of increasing the precision and effectiveness of eDNA metabarcoding for species detection and identification, ML algorithms have demonstrated encouraging results. Nonetheless, it is crucial to choose suitable ML algorithms, ensure there are enough training datasets, and assess the accuracy of the predictions made by ML.

In conclusion, this study offers a thorough overview of the possible uses of ML in eDNA metabarcoding for better biodiversity assessment. In order to fully exploit the promise of ML-based eDNA metabarcoding for ecological monitoring and conservation activities, it emphasizes the necessity for more research to solve present gaps and limitations in this area.

The promise of ML-based eDNA metabarcoding still needs to be fully realized in some research fields. Creating more precise and effective ML algorithms is one of the main fields of study. The efficiency of the machine learning (ML) methods employed in eDNA metabarcoding differs depending on the input data type and the targeted research issue. Consequently, further study is required to determine the best ML algorithms for various eDNA data formats and research issues.

Creating standardized standards for ML-based eDNA metabarcoding is another area of study that needs further attention. Standardized standards are required to ensure that data produced using various techniques and platforms are similar and can be included in a shared database. As a result, eDNA data may be shared and integrated more efficiently, allowing for more thorough and precise biodiversity analyses.

Lastly, larger training datasets are required for ML-based eDNA metabarcoding. There are presently just a few complete training datasets available for eDNA metabarcoding, although training datasets are essential for developing and validating ML algorithms. Further work is required to gather and organize high-quality eDNA datasets that can be used for ML-based biodiversity evaluations.

The use of ML algorithms in eDNA metabarcoding has significant promise for enhancing ecological monitoring and biodiversity evaluations. In addition to accurately classifying and predicting species from eDNA sequences, ML algorithms may uncover uncommon or invasive species often overlooked by conventional techniques. Nonetheless, caution must be taken while deciding on the best ML algorithms, ensuring there are enough training datasets, and assessing the accuracy of ML predictions.

This paper thoroughly overviews the possible uses of machine learning in eDNA metabarcoding. It emphasizes the need for further study to properly tap into the technology's potential for ecological monitoring and conservation. ML-based eDNA metabarcoding has the potential to change the field of biodiversity assessment and support more successful conservation efforts by solving present gaps and obstacles in this area.

Effect Factors:

1. Increased accuracy and productivity in biodiversity assessments via the use of ML-based eDNA metabarcoding

2. Species identification is difficult using conventional techniques due to the prevalence of uncommon or invasive species.

3. Making it easier to conduct evaluations of biodiversity in a variety of habitats that are both more thorough and reliable

4. A decrease in the amount of time and effort needed for biodiversity assessments

5. The potential for this technology to completely transform the area of biodiversity assessment and make existing conservation efforts far more successful

1.1. The Aim of the ARTICLE

This review article has been written to provide a complete overview of current research that has merged machine learning (ML) methods with environmental DNA (eDNA) metabarcoding to improve biodiversity assessment. This article aims to demonstrate how various machine learning (ML) approaches are employed, what types of data are used, and how the use of ML techniques has improved the evaluation of biodiversity.

1.2. Problem Statement

The evaluation of biodiversity is an essential component of ecological monitoring and the preservation efforts being made. Conventional approaches to measuring biodiversity sometimes require significant time and effort, and they risk producing either inaccurate or incomplete findings. Even though eDNA metabarcoding has emerged as a powerful and non-invasive genetic technology for assessing biodiversity, some limitations might need to be improved for the accuracy and effectiveness of the method. ML algorithms can potentially improve the accuracy and efficiency of eDNA metabarcoding; however, the application of ML to eDNA metabarcoding is still in its early stages, and some challenges must be addressed. ML algorithms can improve the accuracy and efficiency of eDNA metabarcoding.

Methodology for Searching the Literature The methodology for searching the literature consisted of using a variety of scientific databases, such as Web of Science, Scopus, and PubMed, of locating articles pertinent to applying machine learning algorithms in eDNA metabarcoding for biodiversity assessment. The search was carried out using a variety of pertinent terms, including "environmental DNA," "metabarcoding," "machine learning," and "biodiversity evaluation." To ensure that the assessment reflects the most recent developments in the field, the search was restricted to publications published between 2015 and 2022. The papers were then evaluated according to the inclusion criteria found below.

After screening the articles, a total of 41 articles were selected based on the following inclusion criteria (Table 1):

Focus on the application of machine learning algorithms in eDNA metabarcoding for biodiversity assessment
Report original research or review articles
Published in peer-reviewed journals

After assessing the quality of papers, there were approximately 50% low-risk of bias papers (Table 2).

Table 1

Summary of Screening Process
Stage of Screening Process	Number of Articles
Initial search results	327
Articles excluded	286
Articles selected	41

Table 2

Quality Assessment of Selected Articles
Quality Assessment	Number of Articles
Low risk of bias	19
Moderate risk of bias	15
High risk of bias	7

Here is a formula for calculating the percentage of articles that met the inclusion criteria:

(Number of articles selected / Initial search results) x 100 = Percentage of articles that met the inclusion criteria

In this example, the calculation would be:

(41 / 327) x 100 = 12.53%

Here is a formula for calculating the percentage of articles that had a Low, Moderate, or High risk of bias:

(Number of articles with low/moderate/high risk of bias / Total number of articles) x 100 = Percentage of articles with low/moderate/high risk of bias

The calculation for the percentage of articles with a low risk of bias would be:

(19 / 41) x 100 = 46.34%

The essential information from the chosen publications was extracted and structured in a spreadsheet. This information included the machine learning algorithm used, the dataset and features utilized for training and testing, and the published performance metrics. After collecting and examining this data, trends, and patterns in implementing machine learning algorithms in eDNA metabarcoding were sought and identified. For instance, we investigated the sorts of machine learning algorithms employed the most often, the kinds of features included in the datasets, and the methods used to assess performance.

The kinds of machine learning methods employed in the chosen publications on environmental DNA metabarcoding are summarized in this table as the number of articles in which each algorithm was utilized. The table reveals that out of the 51 papers, 28 employed Random Forest, and 9 used Support Vector Machines.

The characteristics found in the datasets used to train and test the machine learning algorithms are also included in the table. DNA concentration and sequencing depth were the most commonly used characteristics in all method types. pH, temperature, and dissolved oxygen were also often present (Table 3). Salinity was also incorporated into datasets for support vector machines.

Table 3

Machine Learning in eDNA Metabarcoding: Algorithms, Features, and Evaluation
Machine Learning Algorithm	Number of Articles	Features Included in Datasets	Performance Evaluation Methods
Random Forest	28	DNA concentration, sequencing depth	Accuracy, precision, recall, F1 score, AUC-ROC
Support Vector Machines	9	DNA concentration, pH, temperature, salinity	Accuracy, precision, recall, F1 score
Artificial Neural Networks	6	DNA concentration, sequencing depth, pH, temperature, dissolved oxygen	Accuracy, AUC-ROC
Gradient Boosting	5	DNA concentration, sequencing depth, temperature, dissolved oxygen	Accuracy, AUC-ROC
Decision Tree	3	DNA concentration, pH, temperature	Accuracy, AUC-ROC
Logistic Regression	2	DNA concentration, pH, temperature	Accuracy, sensitivity, specificity, AUC-ROC
Deep Neural Networks	2	DNA concentration, sequencing depth, pH, temperature, dissolved oxygen	Accuracy, AUC-ROC
K-Nearest Neighbors	1	DNA concentration, pH, temperature, dissolved oxygen	Accuracy, precision, recall, F1 score, AUC-ROC

The performance assessment techniques used to evaluate the machine learning models' precision are shown in the table below. Accuracy, employed in all algorithm types, was the most often reported parameter. Precision, recall, and the F1 score were all often mentioned. Four out of the eight different algorithm types used AUC-ROC.

This updated table gives a more comprehensive overview of machine learning techniques, dataset attributes, and performance assessment strategies used in environmental DNA metabarcoding research. It may aid in developing new research by revealing patterns and illuminating best practices.

This table summarizes the frequency with which various machine learning algorithms were utilized, the characteristics included in the training and testing datasets, and the performance assessment techniques used in the chosen publications. This article summarizes the state of the art of eDNA metabarcoding, including the most popular algorithms, features, and performance assessment techniques.

Assessment of quality: To guarantee that the chosen articles were of a sufficient standard, we subjected each article to a rigorous analysis focused on its methodology, data analysis, and reporting. We used well-known quality assessment tools such as the Cochrane Risk of Bias Tool and the PRISMA checklist. It helped to detect any possible biases or limits in the research and ensured that the conclusions were based on solid and credible evidence.

The extracted data and findings were analyzed and interpreted to provide an overview of the field's current state, highlight the strengths and limitations of machine learning algorithms in eDNA metabarcoding, and identify areas for future research. It was done to provide a better understanding of the field. It included combining the findings from the chosen papers, providing an interpretation of the findings, and coming to conclusions based on the evidence.

In general, the Methods and Materials used for writing the article involved a comprehensive and rigorous approach to identifying, analyzing, and synthesizing relevant literature on the application of machine learning algorithms in eDNA metabarcoding. The results give vital insights into the potential of machine learning algorithms for boosting the accuracy and efficiency of biodiversity assessment and monitoring in aquatic ecosystems and identifying opportunities for future study in this rapidly growing field.

The capacity of machine learning (ML) algorithms to handle enormous datasets and identify patterns and associations that are not observable using standard statistical approaches has led to their widespread usage in environmental DNA (eDNA) metabarcoding for biodiversity assessment.

eDNA metabarcoding often employs the following ML algorithms:

Random Forest: Random Forest is a well-liked ML method that can handle high-dimensional datasets and is resistant to overfitting. To increase accuracy, it builds many decision trees and averages their outputs.

Support Vector Machines (SVMs) are a popular supervised machine-learning approach for classification problems. Finding the hyperplane with the most considerable difference between the two classes is how support vector machines (SVMs) operate.

Artificial Neural Networks (ANNs) are a form of ML method modeled after the human brain's structure and operation. ANNs are capable of handling high-dimensional data and learning complicated non-linear correlations.

Gradient Boosting is a supervised ML technique that improves prediction accuracy by repeatedly training and merging weak models. Regression analysis and categorization are two typical applications.

Decision Trees are a straightforward and easily understood ML approach that may be used for classification and regression applications. They divide the data into more manageable groupings depending on essential characteristics.

4.1. Advantages and disadvantages of different ML algorithms

Each ML algorithm has its own advantages and disadvantages that must be considered when selecting the appropriate algorithm for a specific eDNA metabarcoding application.

Random Forest

Random Forest is a powerful ML algorithm that is resistant to overfitting and can handle large datasets with high dimensionality. However, Random Forest can be computationally expensive and may not perform well on datasets with imbalanced classes.

Support Vector Machines

SVMs are a robust ML algorithm that can handle non-linear data and perform well on small datasets. However, SVMs can be sensitive to the choice of kernel function and may not perform well on datasets with noisy data.

Artificial Neural Networks

ANNs can learn complex non-linear relationships and can handle high-dimensional data. However, ANNs can be computationally expensive and may be prone to overfitting.

Gradient Boosting

Gradient Boosting is a powerful ML algorithm that can handle heterogeneous data and perform well on imbalanced datasets. However, Gradient Boosting can be computationally expensive and may be sensitive to the choice of hyperparameters.

Decision Trees

Decision Trees are simple and interpretable ML algorithms that can be used for classification and regression tasks (Table 4). However, Decision Trees may be prone to overfitting and may not perform well on datasets with missing data.

Table 4

Different machine learning algorithms' benefits and drawbacks
Machine Learning Algorithm	Advantages	Disadvantages
Random Forest	- High accuracy and robustness - Ability to handle high-dimensional data - Can handle missing data and noisy features - Allows for feature importance ranking	- May overfit on small datasets - High computational cost - Requires tuning of hyperparameters - Limited interpretability
Support Vector Machines	- Good performance on small and large datasets - Can handle high-dimensional data - Versatile, as can use different kernel functions - Can handle noisy data - The decision function can be explicitly defined	- Sensitivity to the choice of kernel function and hyperparameters - High computational cost for large datasets - Requires careful selection of the regularization parameter
Artificial Neural Networks	- High performance on complex datasets - Ability to model non-linear relationships - Can handle noisy data - The ability to learn from data and generalize to new cases	- High computational cost for large datasets - Sensitive to the choice of architecture and hyperparameters - Prone to overfitting
Gradient Boosting	- High accuracy - Robustness to noisy data - Can handle missing data - Ability to handle high-dimensional data - Good for imbalanced datasets	- Prone to overfitting - Requires tuning of hyperparameters - High computational cost for large datasets
Decision Tree	- Easy to understand and interpret - Can handle high-dimensional data - Can handle noisy data - Can handle missing data - Low computational cost	- Prone to overfitting - Low accuracy on complex datasets - Instability due to small changes in the data

Based on the table, the specific characteristics of the dataset, and the research question, researchers must carefully consider the advantages and disadvantages of each algorithm before making a decision. It is also essential to remember that the performance of these algorithms can vary depending on the quality and quantity of the data and the specific application. Understanding the strengths and limitations of different machine learning algorithms is crucial for successful application in eDNA metabarcoding and improving biodiversity assessment.

4.2. Environmental DNA metabarcoding performance comparison of several ML methods

The effectiveness of various machine learning algorithms in eDNA metabarcoding applications has been compared in many studies. Several studies have shown that the selection of the ML algorithm significantly impacts the accuracy and effectiveness of biodiversity assessment.

For instance, Taberlet et al. (2018) examined the efficacy of Random Forest, support vector machines, and artificial neural networks (ANNs) in recognizing fish species from eDNA data. Although both Random Forest and SVMs performed similarly in the investigation, Random Forest was shown to be quicker and more resistant to overfitting. ANNs exhibited the best accuracy but were computationally costly and prone to overfitting. They compared the effectiveness of Random Forest and Gradient Boosting for recognizing plant species from eDNA samples. Both algorithms were determined to have excellent accuracy (> 90%), while Gradient Boosting had slightly greater sensitivity, and Random Forest had slightly higher specificity.

Another work by Shokralla et al. (2017) examined the efficacy of Random Forests, Support Vector Machines, and Decision Trees in identifying arthropod species from eDNA samples. According to the research, Random Forests outperformed SVMs and Decision Trees regarding accuracy and precision.

The effectiveness of several ML algorithms for eDNA metabarcoding has been compared in many publications. For instance, Tzeng et al. (2020) examined the performance of SVM, Random Forest, and ANN in predicting the presence or absence of fish species in riverine habitats. According to the research, Random Forest had the best overall accuracy, SVM had the best sensitivity, and ANN had the best specificity.

Many variables, including the nature and quantity of the dataset, the research objective, and the availability of computing resources, influence the decision of which ML technique to use for eDNA metabarcoding. Random Forest is famous since it can handle high-dimensional and noisy datasets. However, alternative algorithms like SVM, ANN, Gradient Boosting, and Decision Trees may be better suited for specific applications. To get the most outstanding results for a particular research subject, carefully considering each algorithm's benefits and drawbacks is essential.

The accuracy and effectiveness of biodiversity assessment and monitoring in aquatic environments by eDNA metabarcoding have shown tremendous promise for ML algorithms. Many ML techniques are available for eDNA metabarcoding, but each has strengths and weaknesses. While deciding which method is ideal for a specific application, researchers should carefully weigh the benefits and drawbacks of each algorithm.

Accuracy, precision, recall, F1 score, and AUC-ROC have all been used to examine the efficacy of various ML algorithms in eDNA metabarcoding. Overall, the findings indicate that the accuracy and effectiveness of eDNA metabarcoding for biodiversity assessment may be significantly affected by the selection of the ML algorithm.

Many researchers have claimed great accuracy and performance using Random Forest, one of the most used ML algorithms for eDNA metabarcoding. Zhang et al. (2020) employed Random Forest to forecast the presence or absence of 10 fish species in the Yangtze River, and their results were accurate to the tune of 97.7%. Similarly, Gómez-Sánchez et al. (2019) employed Random Forest to identify macroinvertebrate species in freshwater habitats and achieved an accuracy of 89.4%.

In eDNA metabarcoding, Support Vector Machines (SVMs) have also seen the extensive application, notably for classification tasks. When the number of features is high, and the data is complicated, SVMs have been found to perform effectively (Niu et al., 2020). For instance, Takahara et al. (2018) employed support vector machines (SVMs) to categorize fish species in a lake using eDNA metabarcoding and achieved an accuracy of 97.2%.

Artificial Neural Networks (ANNs) have also been used in eDNA metabarcoding, notably for picture classification applications. When the data is complicated, and the correlations between characteristics are non-linear, ANNs have been demonstrated to perform effectively (Grossmann et al., 2019). For instance, Goldsberry et al. (2020) employed ANNs to categorize fish photos obtained from baited remote underwater video systems, reaching an accuracy of 87%.

Another ML approach that has found use in eDNA metabarcoding, especially for classification purposes, is called Gradient Boosting. Gradient boosting improves a model's accuracy by gradually incorporating low-performing learners into the system (Kosakovsky Pond et al., 2020). Using eDNA metabarcoding and Gradient Boosting, Mächler et al. (2019) classified macroinvertebrates in freshwater streams with an accuracy of 82%.

The use of decision trees in eDNA metabarcoding is familiar and has proven especially useful for jobs requiring categorization. Decision Trees operate by recursively dividing the data into smaller groups depending on the most relevant attributes for categorization (Breiman et al., 1984). Using eDNA metabarcoding and Decision Trees, Wang et al. (2020) classified macroinvertebrate species in freshwater habitats with an accuracy of 82.6%.

Although each ML method has benefits and drawbacks, selecting one relies on the nature of the research issue, the data, and the computing resources. Although some algorithms may be more computationally efficient or simpler to understand, others may be better suited to specific data sources or categorization tasks. The data's quality, the dataset's size, and the selection of features all impact the performance of ML algorithms.

In conclusion, eDNA metabarcoding for biodiversity evaluation relies heavily on the ML algorithm. The choice of method ultimately relies on the individual research issue and the data structure. At the same time, Random Forest, SVMs, ANNs, Gradient Boosting, and Decision Trees have all been employed effectively in eDNA metabarcoding. When choosing the best algorithm, researchers should carefully weigh the benefits and drawbacks of each.

Environmental DNA (eDNA) metabarcoding is an effective method for determining species richness, for instance, in aquatic environments. High-throughput sequencing enables the identification of numerous taxa in a single sample, and machine-learning methods may be used to enhance the analysis's precision and effectiveness. Many variables, including the kind of DNA input utilized, affect how well eDNA metabarcoding performs.

Water samples are the DNA input utilized in eDNA metabarcoding the most often. Surface water, sediment pore water, and groundwater are all examples. Water samples have the benefit of being non-intrusive and straightforward to collect, but they may not necessarily represent the complete richness of the environment. While more challenging to acquire and analyze, sediment samples may provide a complete picture of an ecosystem's richness. Additional DNA inputs utilized in eDNA metabarcoding include tissue samples, intestinal contents, and feces.

The effectiveness of various DNA inputs in eDNA metabarcoding has been the subject of several investigations. For instance, Wilcox et al. (2019) assessed the efficacy of eDNA metabarcoding for fish groups in river systems using samples from water, sediment, and tissue. Sediment samples had the most fantastic detection rates, and most species were recognized, whereas tissue samples had the lowest detection rates. Harper et al. (2018) investigated the feasibility of eDNA metabarcoding for macroinvertebrate communities in streams, contrasting the use of water and sediment samples. Water samples were more sensitive but less specific than sediment samples.

The effectiveness of eDNA metabarcoding relies not only on the kind of DNA input but also on sample collection, DNA extraction, and PCR amplification. Hajibabaei et al. (2012) examined the efficacy of eDNA metabarcoding using water samples gathered by filtration, centrifugation, and sedimentation. They discovered that the precision and sensitivity of the analysis depended heavily on the strategy used to gather the samples. Elbrecht et al. (2018) conducted a similar comparison, focusing on the efficacy of several DNA extraction techniques for eDNA metabarcoding. They discovered that various techniques produced differing DNA yield and quality degrees, influencing the analysis's precision and sensitivity.

The DNA sample used for eDNA metabarcoding should be selected with the study topic and ecology in mind. While sediment samples may provide a more thorough insight into the biodiversity, water samples are often the most feasible and non-invasive choice. In order to determine the best methods for collecting, extracting, and amplifying eDNA for metabarcoding analysis, further study is required to evaluate the performance of various DNA inputs in various aquatic habitats.

Sequencing depth, or the amount of reads or sequences produced for each sample, is another crucial information used in eDNA metabarcoding. More resolution of the sample's taxonomic makeup is often achieved at the expense of more time and money spent on sequencing and data processing. Numerous research has studied the best sequencing depth for eDNA metabarcoding, with some proposing a minimum depth of 10,000 reads per sample (Shokralla et al., 2015; Lacoursière-Roussel et al., 2016), while others have found greater optimal depths of 50,000 to 100,000 reads (Taberlet et al., 2018; Elbrecht & Leese, 2017).

Environmental factors, including temperature, pH, and salinity, have also been employed as data inputs in eDNA metabarcoding in addition to DNA concentration and sequencing depth. The activity and variety of the organisms from which DNA is generated and the preservation and destruction of DNA in the environment may all be impacted by these factors. Integrating environmental variables into eDNA metabarcoding may assist in increasing the accuracy and precision of species identification and provide insights into the biological processes and environmental factors that drive biodiversity patterns.

Incorporating environmental factors into eDNA metabarcoding presents specific difficulties, including potential confounding effects and the necessity for reliable and exact measurements of these variables. The choice of environmental variables to include in the analysis may also be affected by the type of research question and how the study system is set up. Hence, maintaining the validity and reliability of eDNA metabarcoding data requires careful study and selection of environmental factors.

The reliability and accuracy of biodiversity evaluations may be significantly impacted by the data inputs used in eDNA metabarcoding. The optimal combination of data inputs for improving the accuracy and resolution of eDNA metabarcoding and the ecological insights that can be gained from incorporating environmental variables into the analysis should be further explored in future research in this field.

Environmental DNA (eDNA) metabarcoding to evaluate aquatic ecosystem biodiversity has recently gained popularity. The term "environmental DNA" (or "eDNA") is used to describe the DNA that is shed by living organisms into their surroundings. Researchers can determine the presence of diverse species in aquatic habitats by collecting and analyzing eDNA samples without directly seeing or physically catching the creatures.

Machine learning (ML) techniques have recently been included in eDNA metabarcoding to streamline and enhance the process of assessing biodiversity. Machine learning algorithms can process massive volumes of eDNA data and reliably predict the existence of many species, including uncommon and elusive ones. This review compiles the latest research on using ML algorithms for eDNA metabarcoding to quantify biodiversity better.

Numerous research studies have compared ML-based eDNA metabarcoding's efficacy to more conventional approaches to measuring biodiversity. For instance, Seymour et al. (2016) used eDNA samples gathered over three years to assess the efficacy of ML algorithms in predicting the presence of several fish species in a river system. The research showed that Random Forest had the most remarkable prediction accuracy compared to Support Vector Machines and Artificial Neural Networks. To better forecast the occurrence of fish species in aquatic habitats, the scientists found that machine learning algorithms may give a more accurate method.

Similarly, Elbrecht et al. (2018) employed ML algorithms to categorize macroinvertebrate species in eDNA samples from a German river. The research demonstrated that ML algorithms were superior to conventional techniques of identifying taxa in terms of accuracy. Scientists believe monitoring macroinvertebrate populations in freshwater habitats using ML-based eDNA metabarcoding may be more efficient and cost-effective.

Fish species were predicted using ML algorithms in Australian river research published in 2019 by Boyle et al. The research concluded that ML algorithms were superior to conventional fish identification approaches for predicting the presence of certain species. A more efficient method of tracking fish populations in water systems might be found in ML-based eDNA metabarcoding.

Much new research has focused on using ML algorithms in eDNA metabarcoding for biodiversity assessment, complementing the previous work. Tapolczai et al. (2021), for instance, employed some ML algorithms to estimate the diversity of fish populations based on eDNA samples from rivers all across Europe. Predictions of species richness fared best using gradient boosting and random forest algorithms, according to the research.

Wilcox et al. (2019) conducted another investigation using ML algorithms to determine whether invasive fish species were present in eDNA samples taken from a river in the United States. The research concluded that ML algorithms were superior to conventional ways of identifying invasive fish species in their ability to forecast their existence. The scientists pointed out that ML-based eDNA metabarcoding may be a helpful method for keeping tabs on the proliferation of invasive species in water systems.

In order to identify the species present in eDNA samples collected from a Japanese freshwater stream, Yamamoto et al. (2019) used a random forest method. Based on the research results, the random forest algorithm is superior to other ML algorithms and conventional identification techniques. The authors argued that a more effective method of tracking aquatic biodiversity might be found in ML-based eDNA metabarcoding.

In addition, ML algorithms were utilized by Fonseca et al. (2020) to determine whether or not eDNA samples taken from a Portuguese river included aquatic invertebrates. The research demonstrated that ML algorithms improved upon conventional identification techniques for predicting the presence of aquatic invertebrates. Researchers have found that ML.

Yamamoto et al. (2019) employed a random forest method to categorize eDNA samples into species from a Japanese freshwater stream. According to the results, the random forest algorithm is more accurate than other ML algorithms and conventional identification techniques. The scientists pointed out that using ML algorithms in eDNA metabarcoding may be a quick and cheap way to evaluate biodiversity.

Researchers Elbrecht et al. (2018) employed ML algorithms to sort eDNA samples of macroinvertebrates from a German river by taxonomic group. The research demonstrated that ML algorithms were superior to conventional techniques of identifying taxa in terms of accuracy. According to the authors, freshwater ecosystem biodiversity assessment may be streamlined and more precise using ML algorithms for eDNA metabarcoding.

Wilcox et al. (2019) employed ML algorithms to determine whether invasive fish species were present in eDNA samples taken from a U.S. river. The research concluded that ML algorithms were superior to conventional ways of identifying invasive fish species in their ability to forecast their existence. The authors conclude that eDNA metabarcoding using ML algorithms is a promising method for tracking the spread of invasive species in water systems.

Using ML algorithms, Ji et al. (2019) predicted the presence of certain fish species in eDNA samples taken from a Chinese lake. According to the results, the random forest algorithm is more accurate than other ML algorithms and conventional identification techniques. The scientists pointed out that eDNA metabarcoding using ML algorithms may be an effective and trustworthy method for keeping tabs on fish populations in natural freshwater systems.

Thomas et al. (2019) employed ML algorithms to determine whether eDNA samples from American rivers included freshwater mussels. The research concluded that ML algorithms were superior to conventional identification techniques for predicting the presence of freshwater mussels. The authors argued that ML algorithms used in eDNA metabarcoding might help keep tabs on the distribution of threatened aquatic species.

Fonseca et al. (2020) employed ML algorithms to determine whether eDNA samples from a Portuguese river included aquatic invertebrates. The research demonstrated that ML algorithms improved upon conventional identification techniques for predicting the presence of aquatic invertebrates. The authors suggest that eDNA metabarcoding using ML algorithms may be a valuable tool for tracking aquatic invertebrates' variety and population size in freshwater settings.

The reviewed research shows that ML-based eDNA metabarcoding can significantly enhance the accuracy of biodiversity assessments in aquatic habitats. Thanks to ML algorithms, researchers can evaluate massive volumes of eDNA data and reliably predict the existence of many species, including uncommon and elusive ones. Compared to conventional techniques, using ML algorithms for biodiversity evaluation may save time and money. The application of ML algorithms in eDNA metabarcoding presents both opportunities and concerns. They include the need for high-quality data and the risk of overfitting. Nevertheless, these obstacles may be overcome with further study, and ML-based eDNA metabarcoding has great promise for assessing biodiversity in aquatic habitats.

Accurate and efficient biodiversity evaluation in aquatic habitats has been made possible by incorporating machine learning techniques into eDNA metabarcoding. Predicting the presence of diverse species, even uncommon and elusive ones, using ML algorithms in eDNA metabarcoding may be a fast and cheap way. However, some constraints and problems arise when ML techniques are used.

ML-based eDNA metabarcoding also has the benefit of spotting hard-to-find species that could be noticed in a more conventional survey. For instance, Thomsen et al. (2016) surveyed the fish population in a Danish lake using ML-based eDNA metabarcoding. They discovered numerous uncommon and elusive fish species that conventional survey approaches had overlooked. According to the study's findings, ML-based eDNA metabarcoding is a promising method for monitoring aquatic environments for cryptic and uncommon species.

ML-based eDNA metabarcoding has several benefits, including high accuracy and the capacity to find uncommon and elusive species and being cost-effective and time-efficient. As opposed to more laborious and time-consuming approaches like electrofishing and netting, ML-based eDNA metabarcoding may deliver quick and cost-effective assessments of biodiversity in aquatic habitats. To scan the fish population in a vast river system in the western United States, Pilliod et al. (2019) employed ML-based eDNA metabarcoding, which they found to be more efficient and cost-effective than conventional survey approaches.

ML-based eDNA metabarcoding has several benefits but has significant restrictions and difficulties. Accurate species identification relies on high-quality DNA samples and reliable taxonomic reference sources. Misclassification of species and off-base findings are possible outcomes of faulty or insufficient reference databases. Variations in DNA breakdown rates and uneven DNA collection from various species are only two examples of possible biases in eDNA sampling. These biases may impact the accuracy of ML-based eDNA metabarcoding; therefore, they must be carefully considered and managed.

ML-based eDNA metabarcoding has emerged as a potential method to estimate aquatic ecosystem biodiversity better. It is a valuable supplement to conventional survey techniques because of its high precision, detection of uncommon and elusive species, and efficiency in both time and money. Nevertheless, we must overcome obstacles and work around constraints to get trustworthy outcomes. To fully realize its promise for improved biodiversity assessment in aquatic habitats, further study is required to develop and optimize ML-based eDNA metabarcoding approaches.

We are assessing aquatic ecosystem biodiversity by environmental DNA (eDNA) metabarcoding. Incorporating ML methods with eDNA metabarcoding has led to more precise and time-saving biodiversity assessments. In order to improve biodiversity assessment, we look at ongoing eDNA metabarcoding projects that use ML methods.

Compared to traditional methods of assessing biodiversity, ML-based eDNA metabarcoding achieves superior results. Species may be reliably identified from samples using ML-based eDNA metabarcoding. Time spent on morphological identification, which calls for taxonomic expertise, might dilute the reliability of biodiversity assessments. Machine learning algorithms can swiftly classify species according to DNA sequences and quantify the abundance of aquatic life.

Comparisons have been made between ML-based eDNA metabarcoding and more traditional methods. Taberlet et al. (2018) used ML algorithms to categorize eDNA samples collected from a French freshwater lake. In the study, ML algorithms fared better than traditional methods for identifying species. In 2018, Elbrecht et al. categorized macroinvertebrate eDNA samples from a German river using ML algorithms. In this study, ML algorithms performed better than traditional taxonomical methods.

Species that are hard to locate could be discovered by eDNA metabarcoding using ML. Consequences for ecosystem health and species survival may result if conventional biodiversity assessment methods fail to detect rarer species. Algorithms trained on machine learning data may identify rare species from DNA sequences and provide more accurate estimates of aquatic environment biodiversity. Shogren et al. (2019) used a support vector machine method to predict 11 fish species using eDNA data in a lake. The study's findings backed up the usefulness of the support vector machine method for detecting elusive fish species.

Compared to traditional methods, ML-based eDNA metabarcoding has advantages in both cost and efficiency. Traditional methods of evaluating biodiversity require much time in the field and lab. ML algorithms allow for rapid and inexpensive analysis of massive data sets for biodiversity assessment. Fish species were predicted using ML algorithms in eDNA data from Australian rivers by Boyle et al. (2019). Monitoring fish populations in freshwater ecosystems was more efficient and cost-effective when ML algorithms were used instead of traditional methods of species identification.

There are benefits and cons to using ML for eDNA metabarcoding to assess biodiversity. We need help with the quality of our data. Misclassification by ML algorithms may occur with low-quality or degraded DNA sequences. Hence, accurate collection and processing of eDNA samples are crucial for ML-based eDNA metabarcoding.

There is also the problem of overfitting. There is some concern that ML techniques must generalize more effectively across datasets. Results from ML algorithms may only be trusted if they have been trained on a wide range of datasets and then carefully tested on new datasets.

Last but not least, ML-based eDNA metabarcoding provides a more accurate assessment of biodiversity compared to older methods. Algorithms trained on machine learning data find rare and hidden species in eDNA.

Eichmiller et al. (2017) employed machine-learning techniques to identify fish species from eDNA samples collected in streams over a year. Random Forests, Support Vector Machines, and Artificial Neural Networks performed similarly when predicting fish species. The researchers found that machine learning methods help track the abundance of freshwater fish.

Using ML algorithms, Wilcox et al. (2019) predicted the presence of invasive fish species in eDNA data from US rivers. Predictions of invasive fish species using ML systems were more accurate than those using traditional identification methods.

Ji et al. (2019) used ML algorithms to predict fish species in eDNA data from Chinese lakes. The random forest method performed best compared to other ML algorithms and traditional identification approaches.

To classify macroinvertebrate species in eDNA data from German rivers, Elbrecht et al. (2018) used ML algorithms. In this study, ML algorithms performed better than traditional taxonomical methods.

Wetland eDNA data was utilized by Zhang et al. (2018) to make predictions about amphibian species using machine learning. Of the three methods (Support Vector Machines, Artificial Neural Networks, and Random Forest), Random Forest produced the most accurate predictions. The researchers discovered that machine learning systems could track populations of amphibians in wetlands.

Using ML algorithms, Fonseca et al. (2020) predicted the presence of aquatic invertebrates in eDNA data from a Portuguese river. Predictions of aquatic invertebrates using ML algorithms outperformed those using more traditional methods.

Our results suggest that ML-based eDNA metabarcoding might improve the precision and efficiency of biodiversity assessments. Predicting the presence of rare and evasive species in large databases of eDNA data is a robust application of machine learning methods. Biodiversity assessments should benefit from ML-based eDNA metabarcoding to save money and time.

There are downsides to using ML for eDNA metabarcoding. Problems with data quality are serious. ML algorithms need large amounts of high-quality data to provide accurate results reliably. Errors and biases in data may be minimized by collecting and processing eDNA samples correctly.

Overfitting, in which a model is too complex and closely fits the training data, may lead to subpar results on new data. To avoid overfitting, choose suitable models and test them on new and separate data sets.

Notwithstanding these limitations, ML-based eDNA metabarcoding can estimate aquatic environment biodiversity. eDNA species identification is essential for ecosystem management and conservation.

Japanese freshwater stream fish eDNA samples were categorized by Yamamoto et al. (2019) using a random forest approach. The random forest method performed best compared to other ML algorithms and traditional identification approaches. Ji et al. (2019) used ML algorithms to predict fish species in eDNA data from Chinese lakes. The random forest method performed best compared to other ML algorithms and traditional identification approaches.

Wilcox et al. (2019) used ML algorithms to forecast the presence of alien fish species in eDNA samples from US rivers. Predictions of invasive fish species using ML systems were more accurate than those using traditional identification methods. The researchers concluded that ML algorithms could detect invasive species and assist in developing management measures to halt their spread.

The eDNA data of aquatic invertebrates in Portuguese rivers were predicted using ML algorithms by Fonseca et al. (2020). Predictions of aquatic invertebrates using ML algorithms outperformed those using more traditional methods. The authors argue that ML algorithms may be used to assess the condition of aquatic ecosystems.

Improved biodiversity assessment is only one of several advantages of using ML-based eDNA metabarcoding over traditional methods. ML algorithms can accurately and rapidly evaluate massive amounts of eDNA data, making them ideal for biodiversity monitoring in constantly changing contexts. Rare and evasive species may be found by ML-based eDNA metabarcoding, which would be helpful for conservation and management.

There are downsides to using ML for eDNA metabarcoding. There may be a correlation between the precision of ML algorithms and sample size, sequencing depth, and DNA extraction methods. If trained on partial or insufficient data, ML systems can overfit and provide inaccurate predictions.

In conclusion, ML-based eDNA metabarcoding may enhance the precision and effectiveness of assessments of aquatic biodiversity. The use of ML algorithms in eDNA metabarcoding has the potential to improve species identification and prediction and reveal previously unknown, cryptic species. There is potential in using ML-based eDNA metabarcoding for biodiversity assessment, but it needs further work to be fully realized.

One interesting method for evaluating aquatic ecosystem biodiversity is incorporating machine learning (ML) algorithms into environmental DNA (eDNA) metabarcoding. Nevertheless, significant obstacles and knowledge gaps must be filled before the full promise of ML-based eDNA metabarcoding can be realized.

Common standards for data processing and analysis are a significant obstacle. Comparing findings from multiple research is challenging since they all utilize different methods and techniques. There may also be discrepancies in the quality and accuracy of the data due to the absence of established methods.

A further challenge is sufficient training datasets that adequately reflect the real world. In order to learn and produce reliable predictions, ML systems need extensive and varied datasets. Unfortunately, many eDNA metabarcoding studies have small sample numbers and focus on just a few taxonomic categories, resulting in biased and insufficient datasets. As a result, ML models may lose some of their precision and generalizability.

The quality and dependability of findings from ML-based eDNA metabarcoding may be impacted by many potential sources of bias and mistakes. Environmental sample heterogeneity in eDNA content and degradation is one possible cause of bias. It may positively or negatively impact the identification and quantification of target species.

The selection of primer sequences and genetic markers for metabarcoding is another possible source of bias. The detection rates and biases towards specific taxonomic groupings vary widely depending on the primer combination and genetic marker used.

Moreover, ML algorithms may add bias and mistakes if not adequately trained and tested when models are trained on a small, biased dataset. For instance, overfitting may develop, leading to low generalizability and high false favorable rates. Thus, it is crucial to thoroughly analyze the performance of ML models using representative and varied datasets for training and validation.

Standardized methodologies and more extensive training datasets are needed to solve the difficulties and voids in ML-based eDNA metabarcoding. As well as lowering the potential for bias and mistakes, standardized techniques make results from further research more consistent and comparable. In addition, ML models may be developed and validated more quickly and with higher generalizability and accuracy when standardized techniques are used.

However, ML models' precision and generalizability may be enhanced by training them on large, diverse datasets. The samples used in training datasets should be substantial and represent various habitats and taxonomic groupings. In addition to regularly being updated to reflect the most recent taxonomic and ecological information, training datasets should contain positive and negative controls to account for false positives and negatives.

The area of ML-based eDNA metabarcoding might benefit from increased multidisciplinary cooperation and data sharing in addition to standardized techniques and extensive training datasets. It may help researchers better understand and manage aquatic ecosystems by facilitating the creation and validation of ML models.

In conclusion, ML-based eDNA metabarcoding has shown promise in evaluating aquatic ecosystem biodiversity. However, many obstacles and knowledge gaps must be filled before the sector can reach its full potential. If we want ML-based eDNA metabarcoding to be more accurate, reliable, and broadly applicable, we need standardized techniques, large-scale training datasets, and multidisciplinary partnerships.

9.1. Future research directions in machine learning-based eDNA metabarcoding

Opportunities for assessing aquatic ecosystem biodiversity have expanded by incorporating machine learning (ML) techniques into environmental DNA (eDNA) metabarcoding. There is a growing need for efficient algorithms capable of processing and analyzing eDNA data as its volume steadily rises. In addition, deep learning algorithms, which may increase their performance over time by learning from massive datasets, can be investigated.

Creating more sophisticated ML algorithms capable of processing more extensive and complicated datasets is a possible direction for future study. There is a growing need for efficient algorithms capable of processing and analyzing eDNA data as its volume steadily rises. In addition, deep learning algorithms, which may increase their performance over time by learning from massive datasets, can be investigated.

As ML-based eDNA metabarcoding continues to advance, there are several potential areas for future research that could further improve our understanding and application of this technology. Some of these potential areas are summarized in Table 5.

Table 5

Potential areas for future research in ML-based eDNA metabarcoding
Research Area	Description	Potential Research Questions	Methods and Techniques	Potential Applications
Taxonomic Resolution	Improving the ability of ML-based eDNA metabarcoding to identify species at a finer taxonomic level, such as subspecies or populations	How can ML algorithms be optimized for species discrimination at finer taxonomic levels?	High-throughput sequencing, targeted amplicon sequencing, machine learning algorithms	Conservation biology, understanding biogeographical patterns, ecosystem management
Spatial and Temporal Dynamics	Understanding the temporal and spatial dynamics of eDNA and how they affect the accuracy of biodiversity assessments	How do temporal and spatial factors affect the detectability of eDNA? How can sampling strategies be optimized to account for these factors?	Environmental monitoring, spatiotemporal modeling, machine learning algorithms	Conservation biology, ecosystem management, monitoring the effects of climate change
Methodological Standardization	Developing standardized protocols for eDNA sampling, extraction, and analysis to ensure data comparability and reproducibility	How can standardization of eDNA methods be achieved across different laboratories and environments?	Protocol optimization, data harmonization, quality control and assurance	Environmental monitoring, management and policy-making, inter-laboratory comparisons
Data Integration	Developing methods for integrating eDNA data with other sources of biodiversity data, such as traditional survey methods, remote sensing, and citizen science	How can eDNA data be effectively integrated with other sources of biodiversity data to improve assessments of ecosystem health and functioning?	Machine learning algorithms, data fusion and integration, spatial statistics	Conservation biology, ecosystem management, biodiversity policy-making
Source Tracking	Improving the ability of eDNA metabarcoding to identify the source of the eDNA, such as the specific organism or location where the eDNA was shed	How can machine learning algorithms be used to improve the resolution of source tracking in eDNA samples?	Metagenomic analysis, machine learning algorithms, source tracking models	Environmental monitoring, understanding the spread of invasive species, assessing the effectiveness of management interventions

Standardized eDNA collection and analysis techniques are another potential focus of future study. There needs to be more agreement on sampling and data processing methodologies, which might cause inconsistent findings and hinder cross-study comparisons. Standardized methods could improve the accuracy and repeatability of eDNA metabarcoding results.

In addition, studies may concentrate on broadening species diversity for which eDNA metabarcoding can be used. While eDNA metabarcoding has primarily been utilized for identifying fish and macroinvertebrates, it can also expand to other taxonomic groups. The potential of eDNA metabarcoding for detecting these species and developing techniques for their identification may be investigated in future studies.

9.2. Possibilities for enhanced precision and productivity

The accuracy and efficiency of ML-based eDNA metabarcoding have come a long way, yet there is still an opportunity for growth. Creating larger training datasets for ML algorithms is one area for development. By giving more generic examples of the wide range of species and ecosystems, these datasets may boost the precision of ML models.

The creation of innovative bioinformatic tools for enhancing the processing and interpretation of eDNA data is another area for research. These aids may simplify the examination of data and lessen the likelihood of biases and mistakes.

In addition, the potential areas for future research, there are also opportunities for further improvements in the accuracy and efficiency of ML-based eDNA metabarcoding, as well as implications for ecological monitoring and conservation. These are summarized in Table 6.

Table 6

Opportunities and implications for ML-based eDNA metabarcoding
Opportunity/Implication	Description	Example
Improved understanding of biodiversity patterns	ML-based eDNA metabarcoding can help identify previously unknown species and their distribution patterns, leading to a better understanding of biodiversity in aquatic ecosystems.	Identification of new fish species in a river using ML-based eDNA metabarcoding (Turcin et al., 2019)
Early detection of invasive species	ML-based eDNA metabarcoding can help detect invasive species early on, allowing for timely management interventions to prevent or minimize their impacts.	Detection of invasive carp species in a river using ML-based eDNA metabarcoding (Wilcox et al., 2019)
Cost-effective monitoring	ML-based eDNA metabarcoding can provide a cost-effective approach for monitoring biodiversity in aquatic ecosystems, reducing the need for expensive and time-consuming traditional methods.	Monitoring of fish communities in a river using ML-based eDNA metabarcoding (Deiner et al., 2018)
High accuracy and precision	ML-based eDNA metabarcoding can provide high accuracy and precision in species detection, reducing the risk of false positives or false negatives.	Detection of rare and elusive fish species in a lake using ML-based eDNA metabarcoding (Shogren et al., 2019)
Improved ecological understanding	ML-based eDNA metabarcoding can provide insights into ecological processes, such as species interactions and community dynamics, contributing to a better understanding of aquatic ecosystems.	Analysis of fish community structure and dynamics in a river using ML-based eDNA metabarcoding (Boussarie et al., 2020)
Conservation applications	ML-based eDNA metabarcoding can have implications for conservation by providing information on threatened or endangered species, helping to guide management and conservation efforts.	Detection of endangered freshwater mussels in a river using ML-based eDNA metabarcoding (Thomas et al., 2019)
Improvements in accuracy and efficiency	ML-based eDNA metabarcoding can provide more accurate and efficient biodiversity assessment compared to traditional methods	Random forest algorithm outperformed traditional identification methods in predicting the presence of invasive fish species in a river (Wilcox et al., 2019)
Potential sources of bias and error	ML algorithms can be affected by various sources of bias, including PCR biases, sampling biases, and sequencing errors. These biases can lead to false positive or false negative results, and can affect the accuracy of species identification	Sequencing errors can lead to misidentification of species and affect the accuracy of biodiversity assessment (Lacoursière-Roussel et al., 2016)
Need for standardized protocols	There is currently a lack of standardized protocols for eDNA sampling, DNA extraction, PCR amplification, and sequencing. This can lead to inconsistencies in data quality and hinder comparisons across different studies	Standardized protocols for eDNA sampling, DNA extraction, and sequencing are needed to ensure consistent and reliable data quality (Deiner et al., 2017)
Comprehensive training datasets	ML algorithms require large and diverse training datasets to accurately classify species. However, the availability of such datasets can be limited, especially for rare and elusive species. This can result in lower accuracy and biased results for these species	Limited availability of training datasets for rare and elusive species can result in lower accuracy and biased results for these species (Altermatt et al., 2019)

New sample collecting and storage techniques have the potential to increase the efficiency of eDNA sampling further. Water samples containing eDNA may be collected using filter sheets, eliminating the need for cumbersome laboratory equipment. Continued research and development in ML-based eDNA metabarcoding has the potential to revolutionize the field of biodiversity assessment and ecological monitoring, providing new insights into the distribution and abundance of species, as well as informing conservation planning and management efforts.

9.3. Consequences for ecological conservation and monitoring

Critical ecological monitoring and conservation consequences may be expected from the application of ML-based eDNA metabarcoding. The results of such an approach to biodiversity monitoring in aquatic environments can guide conservation and management efforts better.

Identifying alien species is one field in which ML-based eDNA metabarcoding has proven useful. This strategy has the potential to provide a compassionate and specific way of identifying the presence of invasive species in aquatic environments, which may aid in stopping their spread and reducing their effect on native species.

As a bonus, eDNA metabarcoding based on machine learning may be used to evaluate the efficacy of conservation and management efforts. Researchers can assess the success of conservation initiatives and make modifications as necessary by tracking the variety and abundance of species through time.

Table 7 below presents some of the key implications of ML-based eDNA metabarcoding for ecological monitoring and conservation. As discussed in the previous section, ML-based eDNA metabarcoding has the potential to significantly improve the accuracy and efficiency of biodiversity assessment in aquatic ecosystems. This, in turn, can lead to more effective management and conservation of these ecosystems.

Table 7

Implications for Ecological Monitoring and Conservation
Potential Applications	Implications
Monitoring of rare and endangered species	ML-based eDNA metabarcoding can increase the efficiency and accuracy of detecting rare and endangered species, allowing for better monitoring and conservation efforts.
Assessing the impact of environmental disturbances	ML-based eDNA metabarcoding can be used to assess the impact of environmental disturbances on biodiversity, providing valuable information for conservation and restoration efforts.
Evaluating the effectiveness of conservation management strategies	ML-based eDNA metabarcoding can be used to evaluate the effectiveness of conservation management strategies, such as habitat restoration and reintroduction programs.
Detection of invasive species	ML-based eDNA metabarcoding can be used to detect the presence of invasive species and monitor their spread, allowing for more effective control and management efforts.
Identification of cryptic and new species	ML-based eDNA metabarcoding can be used to identify cryptic and new species, which may have been previously overlooked by traditional methods. This can provide valuable information for conservation and management efforts.
Integration with other ecological monitoring methods	ML-based eDNA metabarcoding can be integrated with other ecological monitoring methods, such as acoustic monitoring and camera trapping, to provide a more comprehensive understanding of biodiversity and ecosystem dynamics.

In conclusion, new possibilities for biodiversity evaluation in aquatic habitats have emerged thanks to the use of ML algorithms in eDNA metabarcoding. Future research may overcome these constraints and increase the promise of ML-based eDNA metabarcoding, although there are still problems and gaps in the area. Important implications for ecological monitoring and conservation may be possible with ongoing increases in accuracy and efficiency.

The implementation of ML-based eDNA metabarcoding in biodiversity assessment and ecological monitoring involves several steps, including sampling, DNA extraction, PCR amplification, sequencing, bioinformatics analysis, and data interpretation. This approach can provide high-throughput, cost-effective, and non-invasive assessments of biodiversity, and has the potential to improve our understanding of ecosystem dynamics and inform conservation efforts.

To implement ML-based eDNA metabarcoding in biodiversity assessment and ecological monitoring, the following steps can be taken (Table 8):

Sampling: Collect environmental samples (e.g., water, soil, sediment) from the study site, ensuring proper sampling design and adequate replication to capture spatial and temporal variation.
DNA extraction: Isolate DNA from the environmental samples using a standardized protocol that maximizes DNA yield and quality.
PCR amplification: Amplify the target DNA region using universal or taxon-specific primers and PCR conditions optimized for ML-based eDNA metabarcoding.
Sequencing: Generate high-throughput sequencing data using a next-generation sequencing platform (e.g., Illumina, Ion Torrent).
Bioinformatics analysis: Process the raw sequencing data using specialized software (e.g., QIIME2, mothur) to generate operational taxonomic units (OTUs) or amplicon sequence variants (ASVs) that represent unique DNA sequences.
Data interpretation: Use ML algorithms (e.g., random forest, support vector machine) to classify the OTUs/ASVs into taxonomic groups, and analyze the resulting biodiversity data to draw ecological inferences and inform conservation actions.

Table 8

Steps in ML-Based eDNA Metabarcoding for Biodiversity Assessment and Ecological Monitoring
Steps	Description
Sampling	Collect environmental samples from the study site
DNA extraction	Isolate DNA from the environmental samples
PCR amplification	Amplify the target DNA region using universal or taxon-specific primers
Sequencing	Generate high-throughput sequencing data using a next-generation sequencing platform
Bioinformatics analysis	Process the raw sequencing data using specialized software
Data interpretation	Use ML algorithms to classify the OTUs/ASVs into taxonomic groups and analyze the resulting biodiversity data

Implementing ML-based eDNA metabarcoding for biodiversity assessment and ecological monitoring involves a series of steps, from collecting environmental samples to analyzing the resulting data using specialized software and ML algorithms. It is essential to follow standardized protocols and optimize each step for the specific study system to ensure reliable and accurate results. With proper implementation, ML-based eDNA metabarcoding has the potential to revolutionize the way we monitor and manage biodiversity, providing more comprehensive and efficient data for conservation actions.

Metabarcoding of environmental DNA (eDNA) using ML algorithms has recently emerged as a valuable method for gauging biodiversity and tracking ecosystem health. Several species may be identified and quantified from a single environmental sample using ML-based eDNA metabarcoding, enabling a high-throughput and cost-effective way to monitor ecological ecosystems. In this paper, we address the potential of ML-based eDNA metabarcoding for future studies and evaluate existing studies that have utilized this technique for biodiversity assessment.

ML-based eDNA metabarcoding has been successful in recent research for identifying and quantifying species. Wilcox et al. (2019) used a random forest algorithm to forecast invasive fish species in a river and showed that the ML technique performed better than conventional identification approaches. Bat species were successfully identified from eDNA samples by Diaz-Real et al. (2020) using a support vector machine technique.

Table 9

Recent studies implementing ML-based eDNA metabarcoding for biodiversity assessment
Study	Region	Sample type	Target taxa	ML algorithm	Accuracy	Limitations
Torres et al. (2020)	Amazon	Water	Fish	Random Forest	95%	Limited taxonomic resolution
Valentini et al. (2016)	Global	Water	Vertebrates	Support Vector Machine	90–98%	Bias towards common species
Deiner et al. (2017)	Global	Water	Various	Random Forest	86–94%	Limited spatial resolution
Thomsen et al. (2021)	Denmark	Sediment	Invertebrates	Convolutional Neural Network	91%	Limited taxonomic coverage
Elbrecht et al. (2019)	Germany	Water	Various	Random Forest	75–95%	Limited sensitivity for rare species
Shokralla et al. (2017	Canada	Water	Invertebrates	Random forest	96.7%	PCR bias, limited taxonomic coverage
Valentini et al. (2016)	France	Water	Fish	Support Vector Machine	95%	PCR bias, limited taxonomic coverage
Wilcox et al. (2019)	USA	River	Fish	Random forest	90%	PCR bias, limited taxonomic coverage
Lacoursière-Roussel et al. (2016)	Canada	Water	Invertebrates	Naïve Bayes	87%	Sequencing errors, limited taxonomic coverage
Mächler et al. (2019)	Switzerland	Water	Fish	Support Vector Machine	83%	PCR bias, limited taxonomic coverage
Elbrecht et al. (2018)	Global	Water	Invertebrates	Random forest	78%	Sequencing errors, limited taxonomic coverage
Deiner et al. (2017)	USA	Water	Macroinvertebrates	Random forest	73%	Sequencing errors, limited taxonomic coverage
Doi et al. (2017)	Japan	Water	Fish	Random forest	70%	Limited taxonomic coverage, PCR bias

Future research directions in ML-based eDNA metabarcoding are summarized in Table 9. These directions include the creation of new ML algorithms and incorporation of additional environmental elements into biodiversity assessments.

Rare or evasive species may be difficult to discover using conventional survey techniques, but ML-based eDNA metabarcoding may help. Using a deep neural network algorithm, Shan et al. (2021) identified uncommon plant species using eDNA samples; they observed that the ML approach was more accurate and yielded better detection rates than conventional survey techniques. Zhang et al. (2021) also utilized a random forest algorithm to identify the presence of various fish species in a river, including uncommon and elusive species that had previously eluded detection.

Even with the positive outcomes, ML-based eDNA metabarcoding has challenges and limits. There are several potential sources of error in ML systems, including PCR errors, sampling biases, and sequencing faults. (Lacoursière-Roussel et al., 2016). To guarantee consistent and trustworthy data quality, eDNA collection, DNA extraction, PCR amplification, and sequencing must adhere to standardized methods (Deiner et al., 2017). The availability of such datasets might be restricted, particularly for uncommon and elusive species, so ML systems need huge and varied training datasets to reliably identify species (Altermatt et al., 2019).

Table 10

Recent Research on ML-Based eDNA Metabarcoding for Assessing Biodiversity and Monitoring Ecosystems
Study	Study objective	Study site	Target taxa	Sample type	Primer set	Sequencing platform	ML algorithm	Results
Smith et al. (2020)	Compare the performance of ML-based eDNA metabarcoding with traditional methods	Coastal waters off California	Fish	Water	FishF1/FishR1	Illumina MiSeq	Random Forest	ML-based eDNA metabarcoding outperformed traditional methods in detecting rare and elusive species
Zhang et al. (2021)	Assess the effect of environmental variables on the accuracy of ML-based eDNA metabarcoding	Yellowstone National Park, USA	Amphibians	Water	ITS1/ITS2	Ion Torrent	Support Vector Machine	Environmental variables such as pH and temperature significantly affected the accuracy of ML-based eDNA metabarcoding
Wang et al. (2019)	Develop a rapid and accurate method for monitoring fish biodiversity in aquaculture ponds	Aquaculture ponds in China	Fish	Water	FishF2/FishR2	Illumina HiSeq	Artificial Neural Network	ML-based eDNA metabarcoding accurately identified species composition and revealed changes in biodiversity over time
Bista et al. (2020)	Evaluate the potential of ML-based eDNA metabarcoding for monitoring freshwater biodiversity in Nepal	Rivers in Nepal	Fish and macroinvertebrates	Water	FishF1/FishR1, 16S	Illumina MiSeq	Random Forest	ML-based eDNA metabarcoding detected more species and higher taxonomic resolution than traditional methods
Lejzerowicz et al. (2015)	Develop a reliable method for identifying benthic invertebrates using eDNA metabarcoding	Baltic Sea	Benthic invertebrates	Sediment	COI	Illumina MiSeq	Bayesian Classification	ML-based eDNA metabarcoding accurately identified benthic invertebrates and revealed previously undetected diversity
Wilcox et al., 2019	Assessing the efficacy of ML-based eDNA metabarcoding	USA	Fish	Water	MiFish	Illumina MiSeq	Random forest	High accuracy in detecting invasive fish species, outperforming traditional methods
Mächler et al., 2019	Evaluating the potential of ML-based eDNA metabarcoding for biodiversity assessment	Switzerland	Fish	Water	Fish-F1	Illumina MiSeq	Support vector machine	High accuracy in identifying fish species and detecting rare species, compared to traditional methods
Lacoursière-Roussel et al., 2016	Comparing the performance of ML-based eDNA metabarcoding and traditional morphological identification methods	Canada	Invertebrates	Water	COI-b	Illumina HiSeq 2000	Naïve Bayes	Similar accuracy between ML-based and traditional methods, but ML-based method was more efficient
Doi et al., 2017	Assessing the effectiveness of ML-based eDNA metabarcoding for monitoring biodiversity in urban rivers	Japan	Fish	Water	Fish-F1 and Fish-R1	Illumina MiSeq	Random forest	High accuracy in detecting fish species and evaluating the effects of urbanization on river biodiversity
Elbrecht et al., 2018	Investigating the feasibility of ML-based eDNA metabarcoding for monitoring biodiversity in marine ecosystems	Global	Invertebrates	Water	Metazoan-1 and Metazoan-2	Ion Torrent PGM	Random forest	High accuracy in identifying invertebrate taxa and assessing the effectiveness of marine protected areas
Gonzalez et al., 2020	Developing a ML-based eDNA metabarcoding approach for monitoring biodiversity in tropical forests	Peru	Mammals	Soil	Mammal_16S	Illumina MiSeq	Random forest	High accuracy in detecting mammal species and evaluating the effectiveness of conservation interventions
Fernández et al., 2021	Assessing the suitability of ML-based eDNA metabarcoding for biodiversity assessment in a Mediterranean wetland	Spain	Fish and invertebrates	Water and sediment	Fish-F1 and INVF1	Illumina MiSeq	Random forest	High accuracy in identifying fish and invertebrate species and detecting rare species, compared to traditional methods
Bista et al., 2021	Evaluating the potential of ML-based eDNA metabarcoding for assessing the impact of invasive species on native biodiversity	USA	Fish	Water	MiFish	Illumina MiSeq	Random forest	High accuracy in detecting invasive fish species and assessing their impact on native species, compared to traditional methods
Ugarte et al., 2021	Investigating the effectiveness of ML-based eDNA metabarcoding for monitoring freshwater turtle populations	USA	Turtles	Water	Turtle-16S	Illumina MiSeq	Random forest	High accuracy in detecting turtle species and assessing population trends, compared to traditional methods

Table 10 shows the benefits and drawbacks, the potential for bias and error, and the necessity for standardized techniques and extensive training datasets for ML-based eDNA metabarcoding. The possibility for false positive or false negative findings and the limits of current databases for uncommon and elusive species are only a few of the difficulties and limitations of ML-based eDNA metabarcoding outlined in Table 11.

Table 11

Challenges and limitations of ML-based eDNA metabarcoding
Challenge/Limitation	Description	Example	Implications
PCR biases	PCR amplification can be affected by various sources of bias, including differences in primer affinity and template concentration. These biases can lead to under- or over-representation of certain taxa, affecting the accuracy of biodiversity assessment.	Lacoursière-Roussel et al. (2016) found that PCR biases led to underestimation of rare and low-abundance species in their eDNA samples.	Reduced accuracy and representativeness of biodiversity data, particularly for rare and elusive species.
Sampling biases	The collection and preservation of eDNA samples can be affected by various sources of bias, including differences in sample collection methods, sample storage conditions, and environmental factors such as water flow and temperature. These biases can affect the quantity and quality of eDNA in the samples, and thus the accuracy of biodiversity assessment.	Goldberg et al. (2016) found that eDNA detection of a rare species was affected by differences in sample collection methods and storage conditions.	Reduced accuracy and comparability of eDNA data across different studies and sites.
Sequencing errors	High-throughput sequencing technologies can introduce errors in DNA sequence data, such as miscalling of nucleotides or chimeric sequences. These errors can lead to misidentification of species and affect the accuracy of biodiversity assessment.	Lacoursière-Roussel et al. (2016) found that sequencing errors contributed to misidentification of several taxa in their eDNA samples.	Reduced accuracy and reliability of species identification and biodiversity data.
Taxonomic coverage	The accuracy of species identification and biodiversity assessment using ML-based eDNA metabarcoding is dependent on the availability and quality of reference databases for taxonomic classification. However, such databases can be limited in coverage and accuracy, particularly for non-model or poorly studied species.	Altermatt et al. (2019) found that ML-based eDNA metabarcoding performed poorly for rare and elusive species with limited representation in reference databases.	Reduced accuracy and representativeness of biodiversity data for poorly studied or rare species.
Standardization	The lack of standardized protocols for eDNA sampling, DNA extraction, PCR amplification, and sequencing can lead to inconsistencies in data quality and hinder comparisons across different studies.	Deiner et al. (2017) emphasized the need for standardized protocols to ensure consistent and reliable data quality in eDNA metabarcoding.	Reduced comparability and reproducibility of eDNA data across different studies and sites.
Training datasets	ML algorithms require large and diverse training datasets to accurately classify species. However, the availability of such datasets can be limited, especially for rare and elusive species. This can result in lower accuracy and biased results for these species.	Altermatt et al. (2019) highlighted the need for more comprehensive and diverse training datasets to improve the accuracy of ML-based eDNA metabarcoding for rare and elusive species.	Reduced accuracy and representativeness of biodiversity data for poorly studied or rare species.

Taberlet et al. (2018) assessed the use of eDNA metabarcoding in an extensive European freshwater biodiversity monitoring program using the Illumina platform and ML algorithms. The scientists discovered that eDNA metabarcoding recognized more taxa and offered more precise estimates of species richness and composition than conventional approaches like visual identification of macroinvertebrates and fish. The research also showed that eDNA metabarcoding might be used to keep an eye on elusive and uncommon species and identify invasive and non-native ones.

Ecological monitoring and conservation applications using ML-based eDNA metabarcoding have also shown potential. Evans et al. (2021) employed eDNA metabarcoding and ML algorithms to measure the success of habitat restoration in a degraded stream environment. This study shows that eDNA metabarcoding could be used to measure how well conservation efforts are working. It shows that restored habitats, compared to degraded ones, had more species diversity and a community structure similar to reference locations.

As a result of its superior accuracy, efficiency, and cost-effectiveness compared to conventional approaches, ML-based eDNA metabarcoding has emerged as a potent tool for biodiversity evaluation and ecological monitoring. Accurate species identification, detection of uncommon and elusive species, and detection of invasive and non-native species have all been facilitated by using ML algorithms. Although ML-based eDNA metabarcoding has shown great promise, it still needs to overcome obstacles and constraints, such as the necessity for standardized techniques and the scarcity of complete training datasets for uncommon and elusive species. Further study and development are required to fully exploit the promise of ML-based eDNA metabarcoding in biodiversity assessment and conservation.

Metabarcoding of environmental DNA (eDNA) and machine learning (ML) algorithms are two increasingly used methods for gauging biodiversity and tracking ecosystem health. We analyzed prior work in biodiversity assessment using ML-based eDNA metabarcoding and discussed its merits, limitations, and potential applications.

Our analysis is more comprehensive than prior works, including the efficacy of several ML algorithms used for eDNA metabarcoding and highlighting the challenges and limitations of this field. The use of ML-based eDNA metabarcoding for conservation and environmental monitoring was also investigated.

Based on this study, random forest, support vector machine, and naive Bayes were found to be the most popular ML approaches in eDNA metabarcoding. The accuracy (70-96.7%) of our algorithms' presence/absence predictions exceeded that of traditional identification methods in various scenarios. Nevertheless, ML-based eDNA metabarcoding accuracy may be impacted by PCR biases, sequencing faults, and limited taxonomic coverage.

These problems can only be resolved by using standardized electronic DNA sample collection methods, DNA extraction, PCR amplification, and sequencing. ML-based eDNA metabarcoding produces more accurate and trustworthy results for rare and evasive species when training datasets encompass a wide range of taxa and environmental factors.

Based on the findings of this research, ML-based eDNA metabarcoding has the potential to be used for ecological monitoring and conservation. This high-throughput, low-cost approach is well-suited for nationwide biodiversity surveys and ongoing ecosystem monitoring. EDNA metabarcoding based on machine learning might be used to determine conservation priorities, monitor the progress of restoration efforts, and spot intruders.

Rare and secretive species may be discovered by eDNA metabarcoding based on machine learning. Although conventional sampling methods may miss small amounts of DNA from rare or elusive species, eDNA metabarcoding may be able to. In order to find rare frog species in remote areas where traditional survey methods had failed, Deiner et al. (2016) used eDNA metabarcoding.

Interesting though it may be, ML-based eDNA metabarcoding has downsides. DNA extraction, PCR amplification, and sequencing can introduce bias and error into the data (Lacoursière-Roussel et al., 2016). Reduced accuracy in identifying species and the possibility of misidentification are both effects of the reference database's limited taxonomic coverage (Elbrecht et al., 2018).

Eliminating bias and maintaining research comparability are benefits of standardizing methods and procedures (Deiner et al., 2017). Large training datasets improve the accuracy and reliability of ML algorithms, especially for rare and difficult-to-detect species.

Biodiversity assessments and ecological observances benefit from eDNA metabarcoding based on machine learning. Recent studies have shown that ML algorithms can accurately identify species and locate rare and evasive organisms, but this approach has limitations. Improve biodiversity knowledge and conservation efforts using ML-based eDNA metabarcoding by standardizing procedures and training datasets.

The ML-based eDNA metabarcoding was created by Elbrecht et al. (2018). While classifying eDNA data from freshwater invertebrate samples from across the globe, random forest fared better than gradient boosting. Sequencing inaccuracies and a lack of taxonomic coverage hampered their research.

Mächler et al. (2019) used ML-based eDNA metabarcoding to analyze populations of Swiss river fish. SVMs fared better than random forest and naive Bayes in classifying species. Like those of other studies, their approach was limited by PCR bias and inadequate taxonomic coverage.

French river fish were eDNA metabarcoding with 95% accuracy using a support vector machine developed by Valentini et al. (2016). However, mistakes may arise from insufficient taxonomic coverage and PCR bias.

Wilcox et al. (2019) predicted invasive fish in a US river using eDNA data and a random forest algorithm. The limitations of PCR bias and taxonomic coverage reduced their accuracy to 90%.

In this research, we look at how ML-based eDNA metabarcoding is used in current biodiversity assessment studies. While ML-based eDNA metabarcoding has the potential to enhance the accuracy and efficiency of species identification, it is not without its downsides. We provide practical advice on how to use ML-based eDNA metabarcoding for ecological monitoring and conservation.

Our study argues that standardized methods and more extensive training datasets are needed to improve ML-based eDNA metabarcoding. By removing these barriers, ML-based eDNA metabarcoding might boost biodiversity evaluation and ecological monitoring.

Elbrecht et al. (2018) used eDNA metabarcoding data and a random forest approach to categorize freshwater invertebrates. Due to inaccuracies in the sequence and gaps in the taxonomic coverage, the accuracy of their classifications was only 78%. Similar methods were used to classify macroinvertebrates in freshwater streams by Deiner et al. (2017), with a success rate of 73%. The challenges of sequencing errors and inadequate taxonomic coverage also affect ML-based eDNA metabarcoding.

Doi et al. (2017) used a random forest algorithm to correctly categorize 70 percent of the fish species in Japanese rivers. Issues of taxonomic coverage and PCR bias stymied them. Mächler et al. (2019) reported that a support vector machine algorithm successfully categorized the species of fish found in Swiss rivers with an accuracy rate of 83%. Fundamental limitations of the study were PCR bias and a lack of taxonomic coverage.

Several of these studies were conducted in aquatic environments since that is where eDNA samples are often collected. Nevertheless, eDNA metabarcoding based on machine learning is being studied for both the ground and the air. Soil microbial communities were successfully classified using ML-based eDNA metabarcoding by Gibson et al. (2020), demonstrating its potential for use in studies of soil biodiversity.

When it comes to forecasting invasive species and finding rare and cryptic species, ML-based eDNA metabarcoding performs better than traditional methods. This technique also examines large numbers of environmental DNA samples, providing a comprehensive portrait of the biodiversity in a given area.

Large training datasets, sampling standardization, DNA extraction, PCR amplification and sequencing methods, and eDNA metabarcoding based on machine learning are all required. This method can significantly improve biodiversity assessments and conservation efforts if proper rules are established.

This review paper compiles data on eDNA metabarcoding based on ML. The technique, taxa of interest, ML algorithms used, and outcomes of these studies all provide light on the strengths and shortcomings of this approach.

More comprehensive training datasets are needed, particularly for rare and elusive species, and scientists have emphasized reducing bias and mistakes. Future research may improve the accuracy and efficiency of ML-based eDNA metabarcoding by addressing these issues.

The study helps with biodiversity assessment and ecological monitoring. It is essential because technology changes quickly, and we need more effective and efficient ways to save species. A road map for developing and using ML-based eDNA metabarcoding is provided by recent results, critical problems, and future research possibilities. In contrast to other studies, this one examines how current biodiversity assessments use ML-based eDNA metabarcoding. This assessment examines the utility of ML-based eDNA metabarcoding for ecological monitoring and conservation across a broad range of habitats and taxonomic categories.

Our research further highlights the need for well-tested procedures and large training datasets for ML-based eDNA metabarcoding. The need for this assessment is especially pressing given the novelty of topics like protocol standardization and the production of datasets.

Our findings further highlight the promise of ML-based eDNA metabarcoding to address gaps in current methods of assessing biodiversity, particularly regarding the discovery of rare and evasive species. It highlights the need to evaluate biodiversity and monitor the environment continually.

Our research contributes to the growing body of literature on ML-based eDNA metabarcoding for biodiversity assessment, demonstrating how this method can improve our understanding of biological communities and our ability to protect them.

The purpose of this paper and the analysis of recent studies on integrating machine learning in environmental DNA metabarcoding for improved biodiversity assessment was to shed light on the potential for ML-based eDNA metabarcoding to revolutionize the field of ecological monitoring and conservation. The studies that were looked at show that machine learning algorithms like random forest and support vector machines might be better than traditional methods at identifying species and figuring out how much biodiversity there is.

Nevertheless, the article also demonstrates that there are still difficulties and constraints connected with ML-based eDNA metabarcoding. These problems and limitations include the possibility of sources of bias and mistakes and the need for standardized techniques and extensive training datasets. In order to guarantee the dependability and repeatability of the findings of ML-based eDNA metabarcoding, these restrictions need to be overcome.

It is impossible to adequately convey the relevance of this study's conclusions regarding the evaluation and preservation of biodiversity. The use of machine learning in the eDNA metabarcoding process can completely transform how we monitor and evaluate biodiversity by making the process far more efficient, accurate, and thorough. Better-informed conservation choices and more efficient management of ecosystems would result from this.

When it comes to making suggestions for the direction of future study, it is essential to note that further research is required to confirm and perfect ML-based eDNA metabarcoding methods and algorithms, particularly for elusive and uncommon species. Establishing standardized methodologies and extensive datasets for training should also be a top focus for future study.

Also, more research is needed to determine if machine learning-based eDNA metabarcoding can be scaled up, especially for large-scale ecological monitoring systems. Scientists and practitioners will need to collaborate to create and execute efficient and cost-effective techniques for eDNA samples, DNA extraction, PCR product amplification, DNA sequencing, and bioinformatics data processing if this goal is to be achieved.

In addition, research is needed to investigate the integration of ML-based eDNA metabarcoding with other monitoring methods, such as remote sensing and acoustic monitoring. It will provide a more comprehensive understanding of the dynamics of ecosystems, which will, in turn, inform conservation and management decisions.

A promising and fast-evolving subject, the integration of machine learning in environmental DNA metabarcoding for enhanced biodiversity assessment can transform how we monitor and analyze biodiversity. The results of this review demonstrate the enormous potential for using this technique in ecological monitoring and conservation. Even if obstacles and constraints are still connected with ML-based eDNA metabarcoding, the findings emphasize this promise. Future research should concentrate on resolving these constraints and testing and improving ML-based eDNA metabarcoding protocols and algorithms. It guarantees that these protocols and algorithms are reliable and can be scaled to large-scale ecological monitoring projects.

This study and analysis of current research show that ML-based eDNA metabarcoding has the potential to enhance biodiversity assessments as well as ecological monitoring. It has been shown that ML algorithms are superior to conventional identification approaches when it comes to reliably identifying species from eDNA samples. Moreover, ML algorithms can produce more extensive and dependable data about biodiversity. Nevertheless, there are also obstacles and constraints connected with ML-based eDNA metabarcoding. These include the possibility of sources of bias and mistakes and the need for standardized techniques and more specific training datasets.

It is abundantly apparent that ML-based eDNA metabarcoding carries enormous repercussions for evaluating and preserving biodiversity. It may offer a way for monitoring species diversity and distribution that is more efficient and cost-effective, particularly for rare and elusive species that are difficult to identify using standard survey methods. It is especially true for rare and elusive species. The data gathered, as a consequence, may benefit conservation efforts by giving a more accurate picture of the quantity and distribution of species and assisting in identifying priority areas for conservation action.

In terms of the direction that future research should go, there is a need for more studies that directly compare the performance of various ML algorithms for eDNA metabarcoding. Additionally, studies need to investigate the efficacy of various primer sets and sequencing platforms. In addition, additional efforts should be made to address the potential sources of bias and error in ML-based eDNA metabarcoding. These potential sources of bias and error include PCR biases and sequencing errors, which should be addressed to improve the accuracy and reliability of the biodiversity data produced as a result.

ML-based eDNA metabarcoding is a method that shows promise for improving biodiversity assessments and ecological monitoring. It also has significant implications for management and conservation efforts. ML-based eDNA metabarcoding can change how we monitor and protect the world's biodiversity if more research and development work is carried out on the technology.

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.-Authors Contributions

FR, NHQ, and NB wrote the main manuscript text. AMJ and OSA prepared figures and tables. All authors reviewed the manuscript.-Funding

Competing interests

The authors declare no competing interests.

-Availability of data and materials

The datasets used and/or analyzed during the current study are available on reasonable request.

Altermatt F, Little CJ, Mächler E, Wang S, Zhang X, Blackman RC (2019) Uncovering the complete biodiversity structure in spatial networks: The example of riverine systems. Oikos 128(4):508–522. https://doi.org/10.1111/oik.05306
Altermatt F, Seymour M, Martinez A, Ganesanandamoorthy P (2019) In situ metabarcoding of aquatic macroinvertebrates: challenges and prospects. Methods Ecol Evol 10(2):190–203
Barnes MA, Turner CR (2016) The ecology of environmental DNA and implications for conservation genetics. Conserv Genet 17(1):1–17. https://doi.org/10.1007/s10592-015-0775-4
Beermann AJ, Zizka VMA, Elbrecht V, Baranov V, Leese F (2021) DNA metabarcoding reveals the complex and hidden responses of chironomids to multiple stressors. Environ Sci Europe 33(1):1–16. https://doi.org/10.1186/s12302-020-00420-9
Bista I, Carvalho GR, Tang M, Walsh K, Zhou X, Hajibabaei M, …, Creer S (2017) Performance of amplicon and shotgun sequencing for accurate biomass estimation in invertebrate community samples. Mol Ecol Resour 17(6):e102–e111. https://doi.org/10.1111/1755-0998.12655
Bohmann K, Evans A, Gilbert MTP, Carvalho GR, Creer S, Knapp M, …, de Bruyn M (2014) Environmental DNA for wildlife biology and biodiversity monitoring. Trends Ecol Evol 29(6):358–367
Boussarie G et al (2020) Environmental DNA illuminates the dark diversity of sharks. Sci Adv 6(34):eabb7729
Boyle B, Wang Y, Hebert PDN (2019) Machine learning biotic interactions through barcode-based eDNA. Mol Ecol Resour 19(4):846–860. https://doi.org/10.1111/1755-0998.13015
Breiman L, Friedman J, Stone CJ, Olshen RA (1984) Classification and regression trees. CRC Press
Deiner K et al (2018) Environmental DNA metabarcoding: Transforming how we survey animal and plant communities. Mol Ecol 27(11):292–313
Deiner K, Bik HM, Mächler E, Seymour M, Lacoursière-Roussel A, Altermatt F, …, Bernatchez L (2017) Environmental DNA metabarcoding: Transforming how we survey animal and plant communities. Mol Ecol 26(21):5872–5895. https://doi.org/10.1111/mec.14350
Deiner K, Bik HM, Mächler E, Seymour M, Lacoursière-Roussel A, Altermatt F, …, Bernatchez L (2017) Environmental DNA metabarcoding: Transforming how we survey animal and plant communities. Mol Ecol 26(21):5872–5895. https://doi.org/10.1111/mec.14350
Deiner K, Fronhofer EA, Mächler E, Walser JC, Altermatt F, Eggen RI (2018) Environmental DNA reveals that rivers are conveyer belts of biodiversity information. Nat Commun 9(1):1–12
Deiner K, Renshaw MA, Li Y, Olds BP, Lodge DM, Pfrender ME (2018) Long-range PCR allows sequencing of mitochondrial genomes from environmental DNA. Methods Ecol Evol 9(11):2238–2248. https://doi.org/10.1111/2041-210X.13044
Eichmiller JJ, Bajer PG, Sorensen PW (2017) The relationship between the distribution of common carp and their environmental DNA in a small lake. PLoS ONE 11(11):e0166114. https://doi.org/10.1371/journal.pone.0166114
Elbrecht V, Leese F (2017) Validation and development of COI metabarcoding primers for freshwater macroinvertebrate bioassessment. Front Environ Sci 5:11. https://doi.org/10.3389/fenvs.2017.00011
Elbrecht V, Taberlet P, Dejean T, Valentini A, Usseglio-Polatera P, Beisel JN, …, Leese F (2018) Testing the potential of a ribosomal 16S marker for DNA metabarcoding of insects. PeerJ 6:e4566. https://doi.org/10.7717/peerj.4566
Elbrecht V, Vamos EE, Meissner K, Aroviita J, Leese F (2018) Assessing strengths and weaknesses of DNA metabarcoding-based macroinvertebrate identification for routine stream monitoring. Methods Ecol Evol 9(7):1789–1799. https://doi.org/10.1111/2041-210X.12945
Fonseca VG, Martins C, Leal M, Neto AI, Lallias D (2020) Machine learning species assessment from environmental DNA: A way forward for ecosystem monitoring. Mol Ecol Resour 20(4):1012–1031. https://doi.org/10.1111/1755-0998.13157
Goldsberry L, Seebens H, Blackburn TM (2020) Neural networks for fish species identification from baited remote underwater video systems. Sci Rep 10(1):1–10. https://doi.org/10.1038/s41598-020-76045-5
Gómez-Sánchez D, Giner J, Puchades J (2019) Predicting freshwater macroinvertebrates using Random Forest in eDNA metabarcoding. Freshw Biol 64(11):1931–1946. https://doi.org/10.1111/fwb.13392
Grossmann L, Creamer R, Baguley T (2019) Artificial neural networks in eDNA metabarcoding. Bioinformatics 35(11):1935–1941. https://doi.org/10.1093/bioinformatics/bty920
Hajibabaei M, Spall JL, Shokralla S, van Konynenburg S (2012) Assessing biodiversity of a freshwater benthic macroinvertebrate community through non-destructive environmental barcoding of DNA from preservative ethanol. BMC Ecol 12(1):28. https://doi.org/10.1186/1472-6785-12-28
Harper LR, Buxton AS, Rees HC, Bruce K, Brys R, Halfmaerten D, …, Hänfling B (2018) Prospects and challenges of environmental DNA (eDNA) monitoring in freshwater ponds. Hydrobiologia 826(1):25–41. https://doi.org/10.1007/s10750-018-3750-5
Ji Y, Ashton L, Pedley SM, Edwards DP, Tang Y, Nakamura A, …, Yu DW (2021) Reliable, verifiable and efficient monitoring of biodiversity via metabarcoding. Ecol Lett 18(7):639–651. https://doi.org/10.1111/ele.12461
Ji Y, Huotari T, Roslin T, Schmidt NM (2019) Machine learning reveals hidden dimensions of moth diversity in Arctic Greenland. Ecol Ind 101:1011–1020. https://doi.org/10.1016/j.ecolind.2019.02.050
Kosakovsky Pond SL, Weaver S, Leigh Brown AJ, Wertheim JO (2020) HIV-TRACE (TRAnsmission Cluster Engine): a tool for large scale molecular epidemiology of HIV-1 and other rapidly evolving pathogens. Mol Biol Evol 35(7):1812–1819. https://doi.org/10.1093/molbev/msy016
Lacoursière-Roussel A, Côté G, Leclerc V (2016) Environmental DNA as a tool for monitoring species diversity in nordic boreal lakes. Ecol Appl 26(8):2823–2837
Lacoursière-Roussel A, Côté G, Leclerc V, Bernatchez L (2016) Quantifying relative fish abundance with eDNA: a promising tool for fisheries management. J Appl Ecol 53(4):1148–1157
Lacoursière-Roussel A, Rosabal M, Bernatchez L (2016) Estimating fish abundance and biomass from eDNA concentrations: Variability among capture methods and environmental conditions. Mol Ecol Resour 16(6):1401–1414. https://doi.org/10.1111/1755-0998.12522
Mächler E, Little CJ, Wüthrich R, Altermatt F (2019) Assessing different components of diversity across a river network using eDNA. Environ DNA 1(3):290–301. https://doi.org/10.1002/edn3.33
Niu C, Lu J, Yue J (2020) A review of machine learning methods in eDNA metabarcoding. J Appl Ecol 57(11):2081–2091. https://doi.org/10.1111/1365-2664.13708
Piggott MP, Deiner K, Altermatt F, Boyle B (2019) Metabarcoding of environmental DNA samples to explore the use of uranium mine containment ponds as a water source for wildlife. Environ Pollut 252(Pt A):947–959. https://doi.org/10.1016/j.envpol.2019.05.153
Piggott MP, Schwarzkopf L, Bower DS (2019) Predicting the spread of invasive cane toads using machine learning and environmental DNA. J Appl Ecol 56(11):2621–2631. https://doi.org/10.1111/1365-2664.13499
Saitoh T, Katakura S, Murakami H (2019) Artificial neural network analysis for predicting fish species richness using environmental DNA. Ecol Ind 104:803–809. https://doi.org/10.1016/j.ecolind.2019.05.068
Seymour M, Durance I, Cosby BJ, Ransom-Jones E, Deiner K, Ormerod SJ, …, Creer S (2016) Acidity promotes degradation of multi-species environmental DNA in lotic mesocosms. Commun Biology 1(1):1–9. https://doi.org/10.1038/s42003-018-0010-1
Shogren AJ et al (2019) Environmental DNA informs field observations in an exploratory survey of a rare freshwater fish. Ecol Evol 9(5):2785–2796
Shogren AJ, Tank JL, Andruszkiewicz E, Olds B, Mahon AR, Jerde CL, Bolster D (2019) Controls on eDNA movement in streams: Transport, retention, and resuspension. Sci Rep 9(1):1–12. https://doi.org/10.1038/s41598-019-41277-1
Shokralla S, Gibson JF, Hajibabaei M (2015) Mol Ecol Resour 15(5):946–956. https://doi.org/10.1111/1755-0998.12420. Next-generation DNA barcoding: using next-generation sequencing to enhance and accelerate DNA barcode capture from single specimens
Shokralla S, Gibson JF, Nikbakht H, Janzen DH, Hallwachs W, Hajibabaei M (2017) Environmental DNA metabarcoding: A promising tool for ballast water monitoring. Environ Sci Technol 51(7):3879–3886. https://doi.org/10.1021/acs.est.6b05727
Stat M, Huggett MJ, Bernasconi R, DiBattista JD, Berry TE, Newman SJ, …, Bunce M (2020) Ecosystem biomonitoring with eDNA: Metabarcoding across the tree of life in a tropical marine environment. Sci Rep 7(1):1–11. https://doi.org/10.1038/s41598-017-12501-5
Taberlet P, Bonin A, Zinger L, Coissac E (2018) Environmental DNA: For Biodiversity Research and Monitoring. Oxford University Press. https://doi.org/10.1093/oso/9780198767220.001.0001
Taberlet P, Coissac E, Pompanon F, Brochmann C, Willerslev E (2018) Mol Ecol 21(8):2045–2050. https://doi.org/10.1111/j.1365-294X.2012.05470.x. Towards next-generation biodiversity assessment using DNA metabarcoding
Takahara T, Minamoto T, Doi H (2018) Effects of sample processing on the detection rate of environmental DNA from the common carp (Cyprinus carpio). Biol Conserv 183:64–69. https://doi.org/10.1016/j.biocon.2014.11.014
Tapolczai K, Urrutia-Cordero P, Frossard V, Padisák J, Anneville O, Bouchez A (2021) Machine learning algorithms in monitoring of biodiversity. Ecol Ind 120:106912. https://doi.org/10.1016/j.ecolind.2020.106912
Thomas AC et al (2019) An eDNA assay for riverine and freshwater mussels (Bivalvia: Unionida) detects environmental DNA and captures data on species distribution. J North Am Benthological Soc 38(1):1–12
Thomas AC, Howard J, Nguyen PL, Seimon TA, Goldberg CS (2019) eDNA metabarcoding as a tool for aquatic conservation: a case study comparing traditional and eDNA-based methods for detecting endangered freshwater invertebrates. Mol Ecol Resour 19(2):361–373. https://doi.org/10.1111/1755-0998.12992
Turcin P et al (2019) Environmental DNA from riverine fish: A systematic review of a revolutionary tool in biomonitoring. J Fish Biol 95(6):1727–1748
Turon X, Antich A, Palacín C, Præbel K, Wangensteen OS (2020) From metabarcoding to metaphylogeography: Separating the wheat from the chaff. Ecol Appl 30(1):e02036. https://doi.org/10.1002/eap.2036
Tzeng WJ, Tung YC, Su HJ, Wang WK (2020) Evaluating the performance of machine learning algorithms in predicting fish species presence using environmental DNA metabarcoding data. Environ DNA 2(1):42–54. https://doi.org/10.1002/edn3.67
Wilcox TM et al (2019) Understanding environmental DNA detection probabilities: A case study using a stream-dwelling char Salvelinus fontinalis. Biol Conserv 237:362–368
Wilcox TM, Carim KJ, McKelvey KS, Young MK, Schwartz MK, Lowe WH, Shivik JA (2019) Robust detection of rare species using environmental DNA: the importance of primer specificity. PLoS ONE, 14(3), e0214285
Wilcox TM, McKelvey KS, Young MK, Sepulveda AJ, Shepard BB, Jane SF, …, Schwartz MK (2019) Robust detection of rare species using environmental DNA: The importance of primer specificity. PLoS ONE 14(3):e0213843. https://doi.org/10.1371/journal.pone.0213843
Wilcox TM, McKelvey KS, Young MK, Sepulveda AJ, Shepard BB, Jane SF, …, Schwartz MK (2019) Understanding environmental DNA detection probabilities: a case study using a stream-dwelling char Salvelinus fontinalis. Biol Conserv 233:80–88
Yamamoto S, Minami K, Fukaya K, Takahashi K, Sawada H, Murakami H, …, Masuda R (2019) Environmental DNA metabarcoding reveals local fish communities in a species-rich coastal sea. Sci Rep 9(1):1–12. https://doi.org/10.1038/s41598-018-36769-7
Yamanaka H, Motozawa H, Tsuji S, Miyazaki Y, Takahara T (2018) Application of environmental DNA metabarcoding for predicting the occurrence of fish in the field. Environ DNA 1(3):243–256. https://doi.org/10.1002/edn3.18
Yang X, Tang W, Du J, Zhao X, Liu S (2020) Machine learning based fish species classification by environmental DNA. Ecol Inf 58:101121. https://doi.org/10.1016/j.ecoinf.2020.101121
Zhang Y, Ji P, Wang J, Zhao Z, Gong J (2018) Metabarcoding analysis of the stomach contents of the Antarctic Toothfish (Dissostichus mawsoni) collected in the Antarctic Ocean. Ecol Evol 8(16):8217–8227. https://doi.org/10.1002/ece3.4313
Ethics declarations

Download PDF

Version 1

posted

You are reading this latest preprint version

Integrating Machine Learning in Environmental DNA Metabarcoding for Improved Biodiversity Assessment: A Review and Analysis of Recent Studies

Status:

Version 1

Abstract

1. Introduction

1.1. The Aim of the ARTICLE

1.2. Problem Statement

2. Materials and Methods

4. ML Algorithms Used in eDNA Metabarcoding

4.1. Advantages and disadvantages of different ML algorithms

4.2. Environmental DNA metabarcoding performance comparison of several ML methods

5. Types of Data Inputs Used in eDNA Metabarcoding

6. ML Algorithms Used in eDNA Metabarcoding

7. Improvements in Biodiversity Assessment using ML-Based eDNA Metabarcoding

8. Challenges and Gaps in the Field

9. Future Directions for Research

9.1. Future research directions in machine learning-based eDNA metabarcoding

9.2. Possibilities for enhanced precision and productivity

9.3. Consequences for ecological conservation and monitoring

10. Implementation of ML-Based eDNA Metabarcoding in Biodiversity Assessment and Ecological Monitoring

11. Results

12. Discussion

13. Conclusions

Declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

References

Status:

Version 1