Background: Current studies show early interventions of autism increase significant long-term positive effects, symptoms and, later skills. Currently, These interventions are based on the use of an early diagnostic test. Existing methods for diagnosing Autism Spectrum Disorders (ASDs) such as cognitive tests, Intelligence Quotient, and standardized tests like the Autism Diagnostic Observation Schedule (ADOS) are functionally limited since they rely on child development for diagnoses. The standard is that a child must be at least three(3) years to undergo these tests. Accurate diagnosis is only possible after this period, and this may contribute to delayed diagnosis with an overall effect on the health system. In this era of increasing genetic data, it is possible to infer the genetic patterns of the disorder. This study introduces a novel and rigorous approach for predicting ASDs in neonates and their subsequent severity by identifying significant genes that contribute to the disorder.
Methods: We used a wavelet transform and t-test to identify the significant genes that contribute to the disease. We subsequently employed the Naive Bayes classifier in the prediction of the autistic status of the neonate. Additionally, Principal Component Analysis (PCA) was employed to remove all the dependencies among the genes to enhance classification. Finally, we used the K-means clustering method to predict the severity level of the disease in the neonate.
Results: Up to 200 differentially expressed genes were identified and used for predicting the ASD status of the child with a classification accuracy of 95.91%. Also, the results of the K-means demonstrated that the higher the mean of the cluster, the more severe the disease would be among that corresponding group. Optimizing and implementing these models in clinical settings may significantly reduce the health burden of ASDs.