Correspondence between functional scores from deep mutational scans and predicted effects on protein stability

Abstract

Many methodologically diverse computational methods have been applied to the growing challenge of predicting and interpreting the effects of protein variants. As many pathogenic mutations have a perturbing effect on protein stability or intermolecular interactions, one highly interpretable approach is to use protein structural information to model the physical impacts of variants and predict their likely effects on protein stability and interactions. Previous efforts have assessed the accuracy of stability predictors in reproducing thermodynamically accurate values and evaluated their ability to distinguish between known pathogenic and benign mutations. Here, we take an alternate approach, and explore how well stability predictor scores correlate with functional impacts derived from deep mutational scanning (DMS) experiments. In this work, we compare the predictions of 9 protein stability-based tools against mutant protein fitness values from 45 independent DMS datasets, covering 161,441 unique single amino acid variants. We find that FoldX and Rosetta show the strongest correlations with DMS-based functional scores, similar to their previous top performance in distinguishing between pathogenic and benign variants. For both methods, performance is considerably improved when considering intermolecular interactions from protein complex structures, when available. Finally, we also highlight that predicted stability effects show consistently higher correlations with certain DMS experimental phenotypes, particularly those based upon protein abundance, and, in certain cases, can be competitive with other sequence-based variant effect prediction methodologies for predicting functional scores from DMS experiments.

Protein structural context of cancer mutations reveals molecular mechanisms and identifies novel candidate driver genes

“…Finally, we modelled the effects on protein stability by using FoldX 45 to calculate ΔΔG values for all mutations. Previous work has demonstrated that FoldX outperforms other stability predictors in the identification of disease mutations 46 and shows higher correlations with deep mutational scanning data 47 , providing insights into the molecular mechanisms underlying pathogenic and cancer-associated mutations 23,28 . Because FoldX tends to occasionally output extreme outlier ΔΔG values, and because different proteins can have different intrinsic propensities for destabilising mutations, we introduced a rank normalised metric we call ΔΔG rank for easier visualisation and comparison between proteins.…”

“…Thus, in the absence of any selection, an average ΔΔG rank value of around 0.5 would be expected for a set of random mutations. For the PDB structures, we compute the ΔΔG rank using full complex structures, when available, as the inclusion of intermolecular interactions considerably improves the explanatory value of ΔΔG values 23,47 . For AlphaFold models, we can only evaluate the impact of missense mutations in protein monomers.…”

Updated benchmarking of variant effect predictors using deep mutational scanning

“…A summary of the new VEPs assessed in our benchmark along with their sources is provided in Table 2 , while the full list of VEPs is available in Table EV4 . We did not include any methods focussed on predicting the effects of variants on protein stability, but several of these have been assessed in a recent study (preprint: Gerasimavicius et al , 2023 ).…”

A joint embedding of protein sequence and structure enables robust variant effect predictions

“…For the MSA-based model, we selected GEMME ( Laine et al, 2019 ), which has been shown to produce state-of-the-art results for protein activity prediction outperforming most current machine learning methods using a relatively simple evolutionary model ( Notin et al, 2022a ). For the structure-based model, we selected stability predictions using Rosetta ( Park et al, 2016 ), which are commonly used within protein engineering and have been shown to make useful predictions of protein stability and abundance ( Frenz et al, 2020; Høie et al, 2022; Gerasimavicius et al, 2023 ).…”