PandoraGAN: Generating Antiviral Peptides Using Generative Adversarial Network

The continuous increase in pathogenic viruses and the intensive laboratory research emphasizes the need for cost- and time-efficient drug development. This accelerates research for alternate drug candidates like antiviral peptides (AVP) that have therapeutic and prophylactic potential and gaining attention in recent times. However, diversity in their sequences, limited and non-uniform characterization often limit their applications. Isolating newer peptide backbones with required characteristics is a cumbersome process with many design–test–build cycles. Advanced deep learning approaches such as generative adversarial networks (GAN) can be helpful to expedite the initial stage of developing novel peptide drugs. In this study, we developed PandoraGAN that uses a manually curated training dataset of 130 highly active peptides that include peptides from known databases (such as AVPdb) and literature to generate novel antiviral peptides. The underlying architecture in PandoraGAN is able to learn a good representation of the implicit properties of antiviral peptides. The generated sequences from PandoraGAN are validated based on physico-chemical properties. They are also compared with the training dataset statistically using Pearson’s correlation and Mann–Whitney U-test. We, therefore, confirm that PandoraGAN is capable of generating a novel antiviral peptide backbone showing similar properties to that of the known highly active antiviral peptides. This approach exhibits a potential to discover novel patterns of AVP which may have not been seen earlier with traditional methods. To our knowledge, this is the first ever use of GAN models for antiviral peptides across the viral spectrum.