COVID-19 pandemic defined a worldwide health crisis into a humanitarian crisis. Amid this global emergency, human civilization is under enormous strain since no proper therapeutic method is discovered yet. A wave of research effort has been put towards the invention of therapeutics and vaccines against COVID-19. Contrarily, the spread of this fatal virus has already infected millions of people and claimed many lives all over the world. Computational biology can attempt to understand the protein-protein interactions between the viral protein and host protein. Therefore potential viral-host protein interactions can be identified which is known as crucial information towards the discovery of drugs. In this paper, we have presented an approach for predicting novel interactions from maximal biclusters. Additionally, the predicted interactions are verified from biological perspectives. For this, we conduct a study on the gene ontology and KEGG pathway in relation to the newly predicted interactions.
Research Article
Finding Prediction of Interaction between SARS-CoV-2 and Human Protein: A Data Driven Approach
https://doi.org/10.21203/rs.3.rs-73045/v1
This work is licensed under a CC BY 4.0 License
Journal Publication
published 04 May, 2021
Read the published version in Journal of The Institution of Engineers (India): Series B →
Version 1
posted
You are reading this latest preprint version
Protein-protein interactions (PPI) denote a complex network of reactions that take the responsibility to synchronize and execute the biological process at the cell level in all organisms. PPI dataset is of immense importance in molecular and system biology. Due to its power of revealing the infection mechanism by viral protein, nowadays it plays a crucial role in the study of drug discovery. Thus treatment optimization is enhanced. A large number of experimental and computational approaches [11] have attempted to the way of drug discovery following the PPI dataset as the computational approaches can infer the pre- dicted interactions [6, 5]. Hence, analyzing the PPI dataset of SARS-CoV–2 and human would be useful.
Being a computer science researcher, we could show an efficient algorithmic approach so that the prediction of the protein-protein interaction could be done in an efficient way. Not only that, we expand our scope by incorporating a way to validate the predictions made. Any interaction dataset is easily represented by binary datasets which are easier to store in a compact form. Hence, we find it worthy to use Bimax [9] biclustering approach that could be used to generate biclusters [4]. Also, a systematic manner can find the association rules [1] from generated biclusters and finally the predicted interactions from the binary PPI dataset.
Multiple efforts have already made towards this direction of drug discov- ery. A very few research [3, 2] have been done specifically in case SARS-CoV–2 or COVID–19 disease. Multiple groups of researchers have studied the HIV- human protein interaction. [5, 7, 8] are a few notable contributions among many. The literature has shown that many such analyses are done over the PPI network following the classification based methodology. But it suffers from subjective judgment which is the underlying feature for the designing of a classifier. Specific to the context, for the PPI dataset, a classifier needs both positive and negative interaction data. Though positive cases are validated ex- perimentally, there is no evidence of negative interactions. It is true that based on the quantity of positive data, classifier become reliable. At the same time, the selection of negative data is also important for performance consideration. Thus a good classifier needs proper exploitation of both positive and negative interaction data. This kind of barrier directs us to follow the approach of rule mining based methodology which has been successfully implemented in [5].
So, summarizing the above discussion, in this paper we have followed bi- max biclustering approach on the manually curated protein protein interaction dataset of SARS-COV–2 and human protein. We have shown the procedure for predicting interactions from biclusters and from the generated rules as well. Moreover, we relate our prediction to gene-ontology and KEGG pathway- based study using two well-known bioinformatics tool.
The organization of the remaining part of the paper is as follows. The de- tailed explanation of the followed approach is featured in section 2. In section 3, the description of SARS-CoV–2-Human PPI dataset is presented. Section 4 highlights the predicted interactions and briefly justify their biological rele- vance. Finally, section 5 concludes this paper.
In order to predict new interactions from the existing interaction set, we make use of the association rule mining [1] approach. It is consisting of two main sub-parts. These are finding frequent item-sets (FI) or patterns (FP) and gen- erating association rules from those patterns. As finding FP is computationally expensive, we decide to focus only on the maximal frequent patterns [9]. All the supersets of such patterns are infrequent but information loss arises in the case of the subsets. This is because all subsets are frequent but supporting ob- jects are not available. To combat this information loss while minimizing the computational cost, frequent closed itemset [5] (FCI)/ bicluster mining could be the most favorable way. A FI is called closed if there exists no superset S of FI such that supporting objects of S = supporting objects of FI. Thus using the set of FCI, all FI across the column attributes along with their support- ing row objects can be extracted. In consequence, this approach reduces the number of rules to be presented to the users without losing any information.
In this regard, it can be concluded that finding maximal biclusters from a binary dataset is synonymous to retrieving the FCI in frequent pattern mining problem. Thus the first subproblem is converted into the problem of maximal bicluster generation. The step of generation of maximal biclusters is followed by the step of rule generation and new interaction prediction. These two are the main subtasks here as shown in Figure 1 and are discussed below.
Biclustering [4] is a way to compute the cohesiveness among the elements of a dataset. Bimax [9] is a well-known biclustering algorithm for extracting the maximal biclusters from the binary datasets. It follows a simple divide-conquer approach and extracts all inclusion maximal biclusters, basically, submatrices of all 1s’. An inclusion maximal bicluster indicates that the bicluster is not contained completely in any other bicluster. Here we use bimax as it accom- plishes results within comparable execution time and memory usage along with the other best-performing algorithms. We use the obtained biclusters to pre- dict new interactions and further to generate rules. Additionally, these rules also do prediction for new interactions. A straight forward approach is used to generate the rules from the generated biclusters or maximal closed frequent pattern as explained in [5]. Pictorial representation of the process is shown in Figure 1. Here, we have considered and generated only a non-redundant set of association rules for the necessary knowledge prediction.
The experimental dataset is created by combining PPI interactions from two different sources viz, BioGRID repository [10] and a research article [2]. Bi- oGRID is a conventional biological repository for protein-protein interaction, genetic interaction, etc. We found very less amount of interaction in the data repository to extract sufficient predicted information. For this, an extra predic- tion in the data preprocessing step is carried out. We predict that interactions may possible by the rule of transitivity. We come across the data of intra- organism SARS-CoV–2 protein interactions. We assume that a SARS-CoV–2 protein A interacts with other SARS-CoV–2 proteins X, Y, Z, and human pro- teins as well. Moreover, X, Y, and Z interact with some other human proteins. Hence, SARS-CoV–2 protein A would also interact with the human proteins that have interaction with X, Y, and Z.
Finally, our dataset consists of 28 SARS-CoV–2 proteins and 346 human proteins. The interaction data covers both inter-organism PPI in between hu- man and SARS-CoV–2, as well as intra-organism PPI of SARS-CoV–2. Before applying the transitive-law for the SARS-CoV–2 proteins interaction data, we found 405 interactions. This prediction gives us a total of 1255 interaction data. In our dataset, 1 represents an interaction between the human protein and viral protein. This binary dataset is treated as an input to the bimax algorithm. The input dataset is given as supplementary material along with this article.
SARS-CoV–2-Human PPI dataset is denoted as HxC where the rows repre- sent the human proteins and the columns represent the viral proteins. In this case, we get the clustered itemsets of SARS-CoV–2 proteins that have similar interaction pattern for a subset of objects of human proteins. We have also used the transposed form of the SARS-CoV–2-Human PPI dataset which is represented by the CxH. The aim of this transposed data is to obtain the clusters along with the human proteins for a subset of objects of SARS-CoV–2 proteins. Here, each bicluster specifies that with a subset of human proteins, a subset of SARS-CoV–2 proteins interacts and vice-versa.
Table 1 Statistics for the generated outputs
|
Dataset |
# Biclusters |
# Rules |
Rules > 0.7 Confidence |
Non-redundant Rules |
# Predicted Interactions |
|
|
HxC OLD |
10 |
1 |
0 |
0 |
0 |
|
|
CxH OLD |
9 |
1 |
0 |
0 |
0 |
|
|
HxC |
34 |
30 |
9 |
5 |
48 |
|
|
CxH |
34 |
20 |
11 |
7 |
8 |
|
We use both HxC and CxH as input to the bimax algorithm and summery of generated biclusters can be found in Table 1. It has listed the interac- tion information for two sets of both of the datasets, i.e before (HxC OLD, CxH OLD) and after(HxC, CxH) applying the rule of transition on the orig- inal known protein protein interaction information. The improved results for the updated datasets can be observed from Table 1.
4.1 Interactions prediction from biclusters
Interactions can be predicted from the obtained list of biclusters. Let us take an example of HxC dataset and consider two obtained biclusters in the form of <IS_CP1 : IS_HP1> and <IS_CP2 : IS_HP2> where <IS_CPi: IS_HPi> represents itemsets denoting ith bicluster consisting of SARS-CoV–2 protein and human protein. To predict new interactions from the biclusters itself, we need to take the advantage of set-difference operator. 
Similarly,
Following this way, we obtain a list of predicted interactions, and Table 2 is showing two sets of examples of such a predicted list.
Table 2 Predicted interactions from biclusters
4.2 Prediction by generating rules from the biclusters
This section establishes the approach that we have already mentioned in section 2. CxH dataset gives rules between human proteins and HxC dataset gives rules between SARS-CoV–2 proteins which are true for corresponding SARS-CoV–2 and human protein, respectively. The obtained rules are filtered out in two steps. First, we filter the rules based upon the support and confidence values. Being a sparse dataset, experimentally we fix the support values to a lower threshold as follows. For the CxH dataset, the minimum support count is kept to 4 with a 70% minimum confidence level. Similarly, for the HxC dataset, the minimum support count is 11 with the same minimum confidence level as 70%. Next, the redundancy is eliminated taking the most general rules. Then the interaction prediction is performed. From the final results, the HxC dataset predicts 48 unique and novel interactions. Similarly, from CxH dataset, 8 unique and novel interactions are predicted. To keep the prediction unambiguous, we discard the interactions with a lower confidence level in case of the similar predicted interactions. The predicted interactions obtained from both the datasets are merged and a PPI network is drawn as shown in Figure 2. We find 3 common interactions and thus finally 53 unique novel interactions are obtained from the dataset. The confidence level for each predicted interaction lies within 85% to 90% in most of the cases followed by 90% to 95% interval. For this, we have shown a histogram in Figure 2 along with the interaction network. It can be seen from the figure that SARS-CoV–2 proteins ORF8, NSP12, and E act as network hub. All SARS-CoV–2 proteins are highlighted in yellow color whereas the human proteins are in pink colour. The edge width representing an interaction is getting wider along with the increasing confi- dence level of the interaction. Edge level is showing the value of confidence level for each interaction.
4.3 Gene Ontology based justification of the predicted interactions
To justify the prediction, we have tried to find relevance with the help of biological interpretation. For this, we make use of DAVID (http://david.abcc.ncifcrf.gov), a freely available online bioinformatics repository that provides functional an- notations for a large set of genes. Among the multiple information extracted from this tool, we opt for gene ontology (GO) based study and KEGG Pathway. From all the three domains (Biological Process, Cellular Component, and Molecular Function) covered by GO, we find the GO terms. Moreover, we ex- tract the non-redundant informative GO terms, based upon the p-values (taken from DAVID generated result), by using another tool REVIGO (http://revigo.irb.hr/). It reveals the outliers from the list of submitted GO-terms via checking the semantic similarity and outputs a sorted list based upon dispensability value for each GO-term. More unique terms are having lesser dispensability.
4.3.1 For the predicted interactions obtained from the biclusters
As shown in Table 2, ORF9B is predicted to have interactions with 24 human proteins. Table 3 is showing GO-terms for verifying the predictions. It can be seen from the molecular function of predicted human proteins that these are related to identical binding activities. Study has shown that, ORF9B also has a functional property of membrane binding.
Table 3 Significant GO terms found in the human proteins that are predicted to interact with SARS-CoV-2 Protein ORF9B
|
GO-id |
Term |
% of Proteins |
P value |
Dispensability |
|
GO Category: Biological Process (P-value < 1E-01, Dispensability< 0.6) |
||||
|
GO:1903955 |
positive regulation of protein targeting to mito- chondrion |
12.5 |
5.11E-3 |
0 |
|
GO:0000184 |
nuclear-transcribed mRNA catabolic process, nonsense-mediated decay |
12.5 |
9.57E-3 |
0.019 |
|
GO:0006413 |
translational initiation |
12.5 |
1.25E-3 |
0.106 |
|
GO:0007265 |
Ras protein signal transduction |
8.33 |
8.40E-2 |
0.183 |
|
GO:0016192 |
vesicle-mediated transport |
12.5 |
1.52E-2 |
0.216 |
|
GO:0006364 |
rRNA processing |
12.5 |
2.89E-2 |
0.335 |
|
GO:0034058 |
endosomal vesicle fusion |
8.33 |
1.12E-2 |
0.347 |
|
GO:0045070 |
positive regulation of viral genome replication |
8.33 |
3.20E-2 |
0.445 |
|
GO:0008333 |
endosome to lysosome transport |
8.33 |
4.76E-2 |
0.453 |
|
GO:0016239 |
positive regulation of macroautophagy |
8.33 |
2.83E-2 |
0.468 |
|
GO:0045727 |
positive regulation of translation |
8.33 |
6.42E-2 |
0.516 |
|
GO Category: Cellular Component (P-value < 1E-01, Dispensability < 0.6) |
||||
|
GO:0010494 |
cytoplasmic stress granule |
12.5 |
9.35E-04 |
0 |
|
GO:0016020 |
membrane |
29.17 |
5.05E-2 |
0 |
|
GO:0005622 |
intracellular |
25 |
2.30E-2 |
0.075 |
|
GO:0030123 |
AP-3 adaptor complex |
8.33 |
1.3E-2 |
0.227 |
|
GO:0005829 |
cytosol |
41.67 |
1.51E-2 |
0.327 |
|
GO:0005765 |
lysosomal membrane |
12.5 |
4.63E-2 |
0.33 |
|
GO:0005730 |
nucleolus |
20.83 |
2.10E-2 |
0.377 |
|
GO:0030897 |
HOPS complex |
8.33 |
1.75E-2 |
0.417 |
|
GO:0031519 |
PcG protein complex |
8.33 |
3.35E-2 |
0.491 |
|
GO:0030529 |
intracellular ribonucleoprotein complex |
16.66 |
6.45E-04 |
0.54 |
|
GO Category: Molecular Function (P-value < 1E-01, Dispensability < 0.05) |
||||
|
GO:0044822 |
poly(A) RNA binding | 54.17 | 1.26E-09 | 0 |
|
GO:0000166 |
nucleotide binding | 25 | 5.61E-05 | 0 |
|
GO:0004004 |
ATP-dependent RNA helicase activity | 12.5 | 2.92E-3 | 0 |
|
GO:0003729 |
mRNA binding | 12.5 | 1.04E-2 | 0 |
| GO:0008494 | translation activator activity | 8.33 | 1.11E-2 | 0 |
| GO:0008143 | poly(A) binding | 8.33 | 1.60E-2 | 0 |
| GO:0005515 | protein binding | 70.83 | 2.10E-2 | 0 |
| GO:0003676 | nucleic acid binding | 20.83 | 3.11E-2 | 0 |
Table 4 Significant GO terms found in the human proteins that are predicted to interact with SARS-CoV-2 Protein ORF8
|
GO-id |
Term |
% of Proteins |
P value |
Dispensability |
|
GO Category: Biological Process (P-value < 1E-01, Dispensability < 0.3) |
||||
|
GO:0000398 |
mRNA splicing, via spliceosome |
14.28 |
2.57E-02 |
0 |
|
GO:0017148 |
negative regulation of translation |
9.52 |
6.36E-02 |
0.205 |
|
GO Category: Cellular Component (P-value < 1E-01, Dispensability < 0.2) |
||||
|
GO:0016020 |
membrane |
33.33 |
1.16E-2 |
0 |
|
GO:0071013 |
catalytic step 2 spliceosome |
9.52 |
8.24E-2 |
0 |
|
GO:0005737 |
cytoplasm |
42.85 |
8.30E-2 |
0.15 |
|
GO Category: Molecular Function (P-value < 1E-01, Dispensability < 0.05) |
||||
| GO:0044822 | poly(A) RNA binding | 33.33 | 1.12E-2 | 0 |
4.3.2 For the predicted interactions obtained from the rules
Among the predicted interactions, SARS-CoV–2 protein ORF8 is found to have maximum predicted interactions with 21 human proteins. Followed by this, 16 and 8 numbers of human proteins have been predicted to interact with SARS-CoV–2 proteins NSP12 and E, respectively. Below, Table 4, 5, and 6 are summing up the informative GO-terms verifying the identical biological activities for viral proteins ORF8, NSP12, and E, respectively.
Table 5 Significant GO terms found in the human proteins that are predicted to interact with SARS-CoV-2 Protein NSP12
|
GO-id |
Term |
% of Proteins |
p value |
Dispensability |
|
GO Category: Biological Process (p-value < 1E-04, Dispensability < 0.05)
|
||||
|
GO:0007077 |
mitotic nuclear envelope disassembly |
35.71 |
2.88E-08 |
0 |
|
GO:0006409 |
tRNA export from nucleus |
28.57 |
1.77E-06 |
0 |
|
GO:0010827 |
regulation of glucose transport |
28.57 |
1.95E-06 |
0 |
|
GO:0075733 |
intracellular transport of virus |
28.57 |
7.39E-06 |
0 |
|
GO:1900034 |
regulation of cellular response to heat |
28.57 |
2.37E-05 |
0 |
|
GO:0031047 |
gene silencing by RNA |
28.57 |
7.66E-05 |
0 |
|
GO:0016925 |
protein sumoylation |
28.57 |
8.96E-05 |
0 |
|
GO Category: Cellular Component(p-value < 1E-01, Dispensability < 0.75) |
||||
|
GO:0044613 |
nuclear pore central transport channel |
21.43 |
3.65E-05 |
0 |
|
GO:0001527 |
microfibril |
14.285 |
7.11E-03 |
0 |
|
GO:0005829 |
cytosol |
42.85 |
7.06E-02 |
0.1 |
|
GO:0031012 |
extracellular matrix |
21.42 |
1.82E-02 |
0.5 |
|
GO:0005643 |
nuclear pore |
14.28 |
5.01E-02 |
0.7 |
|
GO:0005578 |
proteinaceous extracellular matrix |
21.42 |
1.51E-02 |
0.7 |
|
GO category: Molecular Function(p-value <1E-01, Dispensability < 0.05) |
||||
|
GO:0017056 |
structural constituent of nuclear pore |
21.43 |
1.14E-04 |
0 |
|
GO:0005515 |
protein binding |
92.86 |
2.65E-03 |
0 |
|
GO:0005487 |
nucleocytoplasmic transporter activity |
14.29 |
1.68E-02 |
0 |
|
GO:0004386 |
helicase activity |
14.29 |
6.35E-02 |
0 |
| GO:0005178 | integrin binding | 14.29 | 7.80E-02 | 0.04 |
| GO:0030023 |
extracellular matrix constituent conferring elasticity |
14.29 | 3.84E-03 | 0.31 |
| GO:0005201 | extracellular matrix structural constituent | 14.29 | 5.04E-02 | 0.4 |
While the activities are highly cohesive within each table data, each indi- vidual table has a distinct set of functions. Table 4 depicts significance GO terms like membrane, poly(A) RNA binding, etc, indicating the involvement of many human proteins in these. It is an important observation as these proteins are expected to have interaction with ORF8 that plays the main role in host- virus interaction. SARS-CoV–2 protein NSP12 is a multifunctional protein and mainly involved in the transcription and replication of viral RNAs. Table 5 is found to have many such common GO-terms related to RNA functions in the biological process. Similarly, from Table 6, it appears that many human proteins are involved in the molecular function of protein transporter activity. It says that these proteins are highly involved in transporting molecules across biological membrane. Hence, the prediction of the human proteins that inter- act with viral protein E is intuitive as it is a small membrane protein having a major role in the assembly of virions.
Table 6 Significant GO terms found in the human proteins that are predicted to interact with SARS-CoV-2 Protein E
|
GO-id Term |
% of Proteins |
P value |
Dispensability |
|
|
GO Category: Biological Process (P-value < 1E-01, Dispensability< 0.7) |
||||
|
GO:0006626 |
protein targeting to mitochondrion |
42.86 |
2.38E-05 |
0 |
|
GO:0007605 |
sensory perception of sound |
28.57 |
3.12E-02 |
0 |
|
GO:0072321 |
chaperone-mediated protein transport |
28.57 |
1.90E-03 |
0.42 |
|
GO:0015031 |
protein transport |
28.57 |
9.08E-02 |
0.68 |
| GO Category: Cellular Component (P-value < 1E-03, Dispensability < 0.7) | ||||
|
GO:0042719 |
mitochondrial intermembrane space protein transporter complex |
42.86 |
9.03E-07 |
0 |
|
GO:0005739 |
mitochondrion |
71.43 |
3.77E-04 |
0.28 |
|
GO:0005743 |
mitochondrial inner membrane |
71.43 |
4.88E-06 |
0.65 |
|
GO:0042721 |
mitochondrial inner membrane protein insertion complex |
28.57 |
9.87E-04 |
0.68 |
|
GO Category: Molecular Function (P-value < 1E-01, Dispensability < 0.5) |
||||
|
GO:0042803 |
protein homodimerization activity |
42.86 |
2.50E-2 |
0 |
|
GO:0008565 |
protein transporter activity |
28.57 |
2.53E-2 |
0 |
|
GO:0005215 |
transporter activity |
28.57 |
6.97E-2 |
0 |
|
GO:0008270 |
zinc ion binding |
42.86 |
6.0E-2 |
0.06 |
|
GO:0051087 |
chaperone binding |
28.57 |
2.94E-2 |
0.46 |
4.4 Justification on the predicted interactions using KEGG pathway
Along with the GO study, we also examine the KEGG pathway obtained from DAVID tool. KEGG pathway has importance in bioinformatics research in understanding genomes, biological pathways, disease, drugs, etc.
4.4.1 For the predicted interactions obtained from the biclusters
Here we would address the example of KEGG pathway enrichment for ORF9B. The tool reveals that the human proteins that are predicted to have interac- tions with ORF9B are related to the Metabolic pathways, pathways to Measles, Herpes simplex infection. Many of the proteins are related to the B-signaling pathway, RNA degradation, etc.
4.4.2 For the predicted interactions obtained from the rules
From the KEGG Pathway obtained using the tool, it has come by the probable interaction of ORF8 with human protein which indicates that multiple cellular activities may lead to Human T-Cell Leukemia Virus Infection. Human pro- teins that are predicted to interact with ORF8, are involved in the pathway of HTLV-I infection, Apoptosis, Hepatitis C, Epstein-Barr virus infection, In- sulin signaling pathway, etc. For the predicted interactions with NSP12, the pathways are Metabolic pathways, RNA transport, etc. Similarly, SARS-CoV–2 protein E is expected to interact with human protein ALG11, IDE that are found to have involvement in Metabolic pathways, Alzheimer’s disease, etc.
Our main aim here is to help in accelerating the procedure of designing the drug and hence an improved medication for COVID–19. For this, we exploit the simple binary SARS-CoV–2-Human PPI dataset. Formation of biclusters, as- sociation rule generation, and the procedure for discovering novel interactions are shown here. Also, the predicted interactions are interpreted biologically for finding their relevance. It has been seen that multiple human proteins that have predicted interaction with a single viral protein share common biological activities. Our study has left scope for consideration of multiple interaction types. Directions of the interactions (viral to host or host to viral) are also crucial information to be examined and working with.
Competing interests:
The authors declare no competing interests.
- Rakesh Agarwal, Ramakrishnan Srikant, et al. Fast algorithms for mining association rules. In of the 20th VLDB Conference, pages 487–499, 1994.
- David E Gordon, Gwendolyn M Jang, Mehdi Bouhaddou, Jiewei Xu, Kirsten Obernier, Kris M White, Matthew J O’Meara, Veronica V Rezelj, Jeffrey Z Guo, Danielle L Swaney, et al. A SARS-CoV-2 protein interaction map reveals targets for drug repur- posing. Nature, pages 1–13,
- Abdul Arif Khan and Zakir Khan. Covid-2019 associated overexpressed Prevotella proteins mediated host-pathogen interactions and their role in coronavirus outbreak. Bioinformatics,
- Sara C Madeira and Arlindo L Oliveira. Biclustering algorithms for biological data anal- ysis: a IEEE/ACM transactions on computational biology and bioinformatics, 1(1):24–45, 2004.
- Kartick Chandra Mondal, Nicolas Pasquier, Anirban Mukhopadhyay, C´elia da Costa Pereira, Ujjwal Maulik, and Andrea GB Tettamanzi. Prediction of protein interactions on HIV-1-human PPI data using a novel closure-based integrated approach. In International Conference on Bioinformatics Models, Methods and Algorithms, pages 164–173. SciTePress,
- Anirban Mukhopadhyay, Ujjwal Maulik, and Sanghamitra A novel biclustering approach to association rule mining for predicting HIV-1–human protein interactions. PLoS One, 7(4), 2012.
- Debasmita Pal, Anindita Sarkar Mondal, and Kartick Chandra Mondal. Knowledge discovery from HIV-1-human PPIs assimilating interaction keywords. In 2016 Interna- tional Conference on Computer, Electrical & Communication Engineering (ICCECE), pages 1–8. IEEE,
- Debasmita Pal and Kartick Chandra Mondal. Predicting novel interactions from hiv-1- human ppi data integrated with protein signatures and go annotations. International Journal of Bioinformatics Research and Applications (IJBRA),
- Amela Preli´c, Stefan Bleuler, Philip Zimmermann, Anja Wille, Peter Bu¨hlmann, Wil- helm Gruissem, Lars Hennig, Lothar Thiele, and Eckart A systematic compar- ison and evaluation of biclustering methods for gene expression data. Bioinformatics, 22(9):1122–1129, 2006.
- Stark, B. J. Breitkreutz, L. Reguly, T.and Boucher, and M. Breitkreutz, A.and Ty- ers. Biogrid: a general repository for interaction datasets. Nucleic acids research, 34(Database issue):D535—-D539, 2006.
- Hina Umbrin and Saba Latif. A survey on protein protein interactions (ppi) meth- ods, databases, challenges and future directions. In 2018 International Conference on Computing, Mathematics and Engineering Technologies (iCoMET), pages 1–6. IEEE,
- PPICOVIDHUMAN.csv
COVID-Human protein- protein interaction database
- PPICOVIDHUMANT.csv
Transposed COVID-Human protei-protein interaction database
- FinalPredictedInteractions.xlsx
Predicted list of interactions found