PEPTIDE BASED VACCINE GENERATION SYSTEM WITH DUAL PROJECTION GENERATIVE ADVERSARIAL NETWORKS
A computer-implemented method is provided for generating new binding peptides to Major Histocompatibility Complex (MHC) proteins. The method includes training, by a processor device, a Generative Adversarial Network GAN having a generator and a discriminator only on a set of binding peptide sequences given training data comprising the set of binding peptide sequences and a set of non-binding peptide sequences. A GAN training objective includes the discriminator being iteratively updated to distinguish generated peptide sequences from sampled binding peptide sequences as fake or real and the generator being iteratively updated to fool the discriminator. The training includes optimizing the GAN training objective while learning two projection vectors for a binding class with two cross-entropy losses. A first loss discriminating binding peptide sequences in the training data from non-binding peptide sequences in the training data. A second loss discriminating generated binding peptide sequences from non-binding peptide sequences in the training data.
1 . A computer-implemented method for generating new binding peptides to Major Histocompatibility Complex (MHC) proteins, comprising:
transforming, by a generator of a Generative Adversarial Network GAN, a sampled latent code vector from a multivariate unit-variance Gaussian distribution obtained from training data and a sampled binding class label of the training data to a peptide feature representation matrix with each column corresponding to an amino acid, the training data comprising a set of binding peptide sequences and a set of non-binding peptide sequences, the sampled binding class label of the set of binding peptide sequences being different from the sampled binding class label of the set of non-binding peptide sequences.
2 . The computer-implemented method of claim 1 , further comprising:
training, by a processor device, the Generative Adversarial Network GAN having the generator and a discriminator only on the set of binding peptide sequences given the training data, wherein a GAN training objective comprises the discriminator being iteratively updated to distinguish generated peptide sequences from sampled binding peptide sequences as fake or real and the generator being iteratively updated to fool the discriminator,
wherein said training comprises optimizing the GAN training objective while learning two projection vectors for a binding class with two cross-entropy losses, a first of the two cross-entropy losses discriminating binding peptide sequences in the training data from non-binding peptide sequences in the training data, and a second of the two cross-entropy losses discriminating generated binding peptide sequences from non-binding peptide sequences in the training data.
3 . The computer-implemented method of claim 2 , wherein the GAN is a Wasserstein GAN.
4 . The computer-implemented method of claim 2 , wherein said training further comprises updating the generator with tempering Softmax units to minimize the two cross-entropy losses.
5 . The computer-implemented method of claim 4 , wherein the tempering Softmax units are employed with entropy regularization for implicit temperature control of the tempering Softmax units.
6 . The computer-implemented method of claim 2 , wherein the generator is a deep neural network, comprising a convolutional layer for receiving the set of binding peptide sequences.
7 . The computer-implemented method of claim 2 , wherein the GAN operates on the sampled latent code vector from the multivariate Gaussian distribution obtained from the training data.
8 . The computer-implemented method of claim 2 , wherein the two cross-entropy losses are implemented by dual classifiers.
9 . The computer-implemented method of claim 2 , further comprising generating peptide-based vaccines with user-specified properties using the trained GAN.
10 . The computer-implemented method of claim 9 , wherein the peptide-based vaccines are output from the generator as Softmax output units, and wherein the generator comprises a fully-connected layer for receiving an input random noise vector and another fully-connected layer for outputting the Softmax output units.