Protein Design with Deep Learning

Abstract

Computational Protein Design (CPD) has produced impressive results for engineering new proteins, resulting in a wide variety of applications. In the past few years, various efforts have aimed at replacing or improving existing design methods using Deep Learning technology to leverage the amount of publicly available protein data. Deep Learning (DL) is a very powerful tool to extract patterns from raw data, provided that data are formatted as mathematical objects and the architecture processing them is well suited to the targeted problem. In the case of protein data, specific representations are needed for both the amino acid sequence and the protein structure in order to capture respectively 1D and 3D information. As no consensus has been reached about the most suitable representations, this review describes the representations used so far, discusses their strengths and weaknesses, and details their associated DL architecture for design and related tasks.

A Deep SE(3)-Equivariant Model for Learning Inverse Protein Folding

“…We now give a brief overview CNN and GNN-based methods for inverse folding. For a more detailed review of feature representations applied to CPD models, see (Defresne et al 2021).…”

From polymerase engineering to semi-synthetic life: artificial expansion of the central dogma

“…To further expand the central dogma, more genetic polymers with novel modifications or combination of modifications can be designed and synthesized, and their efficient polymerases also need to be discovered or engineered, with the development and employment of novel polymerase evolution strategies, as well as the assistance of computational tools, including novel machine-learning methods 228,229 .…”

Deep learning approaches for conformational flexibility and switching properties in protein design

“…Ultimately, even if proteins with purpose-built conformational flexibility or switchable state can be constructed, any functionality must be verified through further in silico validation ( Nivedha et al, 2018 ; Chen et al, 2020 ), meaning programming function from sequence generation is not end-to-end differentiable. Analogous to structure prediction, structure generation methods inherently account for local side-chain flexibility owing to the ensemble of rotamer positions examined per residue during construction ( Defresne et al, 2021 ). Yet there remains a conceptual gap to more extensive conformational flexibility.…”