Accurate Mutation Effect Prediction using RoseTTAFold

Abstract

Predicting the effects of mutations on protein function is an outstanding challenge. Here we assess the performance of the deep learning based RoseTTAFold structure prediction and design method for unsupervised mutation effect prediction. Using RoseTTAFold in inference mode, without any additional training, we obtain state of the art accuracy on predicting mutation effects for a set of diverse protein families. Thus, although the architecture of RoseTTAFold was developed to address the protein structure prediction problem, during model training RoseTTAFold acquired an understanding of the mutational landscapes of proteins comparable to that of large recently developed language models. The ability to reason over structure as well as sequence could enable even more precise mutation effect predictions following supervised training.

AlphaFold2 can predict single-mutation effects

“…Furthermore, as accuracy improves, evaluating the quality of predictions becomes increasingly complicated by the inherent noise in experimental measurements [16][17][18][19][20][21][22][23]. So far, no study has evaluated whether AF can accurately measure structural changes due to single mutations, and there are conflicting reports as to whether AF can predict the effect of a mutation on protein stability [24][25][26][27][28]. Furthermore, recent evidence suggests that AF learns the energy functional underlying folding, raising the question of whether the inferred functional is sensitive enough to discern the subtle physical changes due to a single mutation [29].…”

AlphaFold2 and Deep Learning for Elucidating Enzyme Conformational Flexibility and Its Application for Design

“…71 RF has also the potential to predict the effect of mutations on protein function. 72 The RF-based diffusion model (named RFdif fusion) has been recently developed by the Baker lab. 73 RFdif f usion can very rapidly and accurately design topology-constrained protein monomers, protein binders, symmetric oligomers, metal-binding proteins, and even enzyme scaffolds containing specific active-site residues.…”

“…Similarly, RF instead of AF2 was used for designing high-affinity protein binders or proteins with prespecified functional motifs . RF has also the potential to predict the effect of mutations on protein function …”

De Novo Design of Polyhedral Protein Assemblies: Before and After the AI Revolution

“…AF-multimer v2 predictions were again run using a local install of ColabFold. [162] made efforts in predicting the effect of potential mutations on protein folding and function, [140][141][142] as well as in the context of protein-protein interfaces. [136] Current design approaches yield polyhedral assemblies with a narrow range of structural properties (e. g. symmetry, diameter, porosity and cavity volume), [23] in contrast to natural self-assembling systems such as clathrin, which can adapt to its cargo and assemble across a variety of architectures, [143,144] and virus-like particles (VLPs), which can not only form icosahedral compartments with more than 60 copies of the assembling subunit by breaking local symmetry, [145,146] but can also be engineered to assemble into more complex architectures.…”