Investigating the impacts of sidechains on de-novo protein design

Abstract

De novo protein design aims to create novel protein structures and sequences, often to enable specific functions. Most current generative models operate on simplified backbone-only representations. However, in vivo and in vitro protein folding and function are largely mediated by amino acid sidechains. Given this biochemical relevance, we ask: what happens when sidechain and sequence information are introduced into a protein generative model? To address this, we developed SiBaSe, a novel generative model that simultaneously co-designs sidechains, backbones, and sequence. SiBaSe achieves design performance approaching that of state-of-the-art backbone-only models. However, analysis of design patterns reveals that, despite access to sidechain data, the model behaves similarly to a backbone-based model. This appears to arise from uncertainties in the simultaneous modeling of sequence and sidechains that are inherent to flow-based architectures and offers new insight into architectural limitations and opportunities for improving generative protein design.