Inpainting Protein Sequence and Structure with ProtFill

Elizaveta Kozlova, Arthur Valentin, Daniel Nakhaee-Zadeh Gutierrez

Keywords: protein design, co-design, diffusion, GNN

TL;DR: We present ProtFill, a new protein sequence and structure co-design model that works on a wide range of proteins and outperforms existing alternatives on sequence prediction.

Abstract: Designing new proteins with specific binding capabilities is a challenging task that has the potential to revolutionize many fields, including medicine and material science. Here we introduce ProtFill, a novel method for the simultaneous design of protein structures and sequences. Employing an $SE(3)$ equivariant diffusion graph neural network, our method excels in both sequence prediction and structure recovery compared to SOTA models. We incorporate edge feature updates in GVP-GNN message passing layers to refine our design process. The model's applicability for the interface redesign task is showcased for antibodies as well as other proteins. The code is available at https://github.com/adaptyvbio/ProtFill.

Submission Number: 86