Classification of Antiviral Peptides Using Amino Acid Based Sequences

Antiviral peptides are short molecules that stop viruses from penetrating the cell and potentially infecting or reproducing inside them. These peptides work in different ways to achieve the result of preventing viruses from penetrating the cell by blocking viruses from entering cells and interfering with the steps of the viral replication cycle. As a result, antiviral peptides are important because they present a new way to fight viral diseases, especially as viruses become more resistant to drugs and new variations emerge. I developed a machine learning model to predict whether a peptide is antiviral by using its amino-acid sequence. I used the k-mer Bag-of-Words model to convert amino-acid sequences into numbers and then trained a Logistic Regression classifier and a neural network on a dataset of antiviral peptides. The best model, which was the Logistic Regression classifier, achieved an area under the ROC curve (AUC) of 0.940, classification accuracy (CA) of 0.932, precision of 0.929, and a recall of 0.932. Thus, these results show that my machine learning model is effective at identifying antiviral peptides, which could help discover new antiviral drugs.