Identificationof Protein–Protein Interaction(PPI) Sites on the Influenza A (H1N1) Viral Genome Using GradientBoosting and Artificial Neural Network (ANN) Models
Identification of protein–protein interaction(PPI) sitesis crucial for understanding molecular recognition. Experimental identificationof PPI is expensive, time-consuming, and laborious. A large numberof computational methods addressed this problem. However, no computationsspecifically addressed PPI site prediction for the frequently mutatinginfluenza A virus (IAV) genome that invades human hosts. For the firsttime, we report the prediction of PPI sites on the IAV genome (proteinsequences). The method was benchmarked across various machine-learningmodels, optimizing class imbalance and unlabeled data types. The best-performingmodels were (i) the gradient boosting model, augmented with minorityclass oversampling and positive unlabeled (PU) learning and (ii) theprotein-specific bidirectional encoder representations from transformers(Prot-BERT) combined with an artificial neural network (ANN) (termedthe Prot-BERT-ANN model), adjusted with class weight correction andthreshold tuning. The models were trained on two types of interactionsite data sets: one obtained from diverse protein families (Train-1)(17995 amino acid sites) with known interaction sites from proteinstructures and the other from the IAV consensus protein sequences(3322 amino acid sites) with experimentally annotated PPI sites onthe conserved regions of the proteins (Train-2). External validationwas performed on two test data sets: (i) from six IAV proteins, M1,NS1, NEP, NP, PB1, and PB2, reported to interact with host factors,with experimentally annotated PPI sites on the nonconserved regionof the proteins (Test-1), and (ii) the SARS-CoV-2 spike protein sequence(195 amino acid sites) (Test-2). Blind prediction was performed onthree IAV protein sequencesNA, HA, and M2curated fromthe Human Viral Interaction Database (HVIDB). The prediction aimedto decipher the effect of amino acid substitutions on the protein–proteininteraction sites of the viral genome. The gradient boosting methodwith oversampling and PU learning, trained on the Train-2 data set,consistently performed better on both external validation data sets.The recall values obtained from the predictions on the Test-1 dataset were compared with the published D-SCRIPT (a neural language-basedmodel) results. The gradient boosting model showed a higher averagerecall value (0.53 ± 0.04) for six IAV proteins compared to theD-SCRIPT results (0.18 ± 0.19). The gradient boosting predictionfor the experimentally reported PPI sites on the SARS-CoV-2 spikeprotein (Test-2 data set) was 55% accurate, despite Test-2 being independentof Train-2. The results indicated the generalizability and interpretabilityof the gradient boosting model for IAV PPI site predictions. The effectsof amino acid substitutions on PPI sites were demonstrated on fiveMatrix 1 (M1) protein sequences. This approach could be used to identifythe PPI sites on newly emerging viral strains (e.g., influenza virus,SARS-CoV-2, etc.) with potential applications for drug design, improvementof drug binding, or drug repurposing, subject to further validation.
Read more