ProteinSGM: Score-based generative modeling for de novo protein design

Abstract

<title>Abstract</title> Score-based generative models are a novel class of generative models that have shown state-of-the-art sample quality in image synthesis, surpassing the performance of GANs in multiple tasks. Here we present ProteinSGM, a score-based generative model that produces realistic <italic>de novo</italic> proteins and can inpaint plausible backbones and functional sites into structures of predefined length. With <italic>unconditional generation</italic>, we show that score-based generative models can generate native-like protein structures, surpassing the performance of previously reported generative models. We apply <italic>conditional generation</italic> to <italic>de novo</italic> protein design by formulating it as an image inpainting problem, allowing precise and modular design of protein structure.