Generating Antimicrobial Peptides from Latent Secondary Structure Space
Antimicrobial peptides (AMPs) have shown promising results in broad-spectrum antibiotics and resistant infection treatments, which makes it attract plenty of attention in drug discovery. Recently, many researchers bring deep generative models to AMP design. However, few studies consider structure information during the generation, though it has shown crucial influence on antimicrobial activity in all AMP mechanism theories. In this paper, we propose LSSAMP that uses the multi-scale VQ-VAE to learn the positional latent spaces modeling the secondary structure. By sampling in the latent secondary structure space, we can generate peptides with ideal amino acids and secondary structures at the same time. Experimental results show that our LSSAMP can generate peptides with multiply ideal physical attributes and a high probability of being predicted as AMPs by public AMP prediction models.
PDF Abstract
