A peptide-based vaccine generation system using dual projection generative adversarial networks

Abstract

A method for generating novel binding peptides for MHC proteins includes training (430) a generative adversarial network (GAN) having only a generator and a discriminator on a set of binding peptide sequences given training data including a set of binding peptide sequences and a set of non-binding peptide sequences by a processor device. Training objectives of the GAN include iteratively updating the discriminator to identify the generated peptide sequences as fake or real from the sampled binding peptide sequences, and iteratively updating the generator to fool the discriminator. The training includes learning two projection vectors for binding classes with two cross-entropy losses while optimizing (440) the training objective of the GAN. The first loss is to discriminate between binding peptide sequences in the training data and non-binding peptide sequences. The second loss is to discriminate between the generated binding peptide sequences and non-binding peptide sequences in the training data.

Images