Accurate protein stability predictions from homology models

Abstract

Calculating changes in protein stability (ΔΔG) has been shown to be central for predicting the consequences of single amino acid substitutions in protein engineering as well as interpretation of genomic variants for disease risk. Structure-based calculations are considered most accurate, however the tools used to calculate ΔΔGs have been developed on experimentally resolved structures. Extending those calculations to homology models based on related proteins would greatly extend their applicability as large parts of e.g. the human proteome are not structurally resolved. In this study we aim to investigate the accuracy of ΔΔG values predicted on homology models compared to crystal structures. Specifically, we identified four proteins with a large number of experimentally tested ΔΔGs and templates for homology modeling across a broad range of sequence identities, and selected three methods for ΔΔG calculations to test. We find that ΔΔG-values predicted from homology models compare equally well to experimental ΔΔGs as those predicted on experimentally established crystal structures, as long as the sequence identity of the model template to the target protein is at least 40%. In particular, the Rosetta cartesian_ddg protocol is robust against the small perturbations in the structure which homology modeling introduces. In an independent assessment, we observe a similar trend when using ΔΔGs to categorize variants as low or wild-type-like abundance. Overall, our results show that stability calculations performed on homology models can substitute for those on crystal structures with acceptable accuracy as long as the model is built on a template with sequence identity of at least 40% to the target protein.

Target-template relationships in protein structure prediction and their effect on the accuracy of thermostability calculations

“…Given a high homology template for the regions of interest in a protein, the absence of a crystal structure is no longer a major roadblock when predicting thermostabilities. The results presented here agree with the conclusions from recent studies where thermostabilities were calculated for models generated by AlphaFold and homology modeling 56,58,59 . To get results consistent with the experiment, the authors suggest using templates with at least 40% sequence identity.…”

Rapid protein stability prediction using deep learning representations

“…To test this, we used template-based (homology) modelling to generate structures of the four proteins selected from ProTherm that we analysed above. To minimize issues of leakage between training and testing data, we have recently used MODELLER ( Martí-Renom et al, 2000 ; Webb and Sali, 2016 ) to construct models of the four proteins, using templates with decreasing sequence identities to the original structures ( Valanciute et al, 2022 ). We used these structures as input to RaSP in order to calculate ΔΔ G values to compare with experiments;we also performed similar tests using ΔΔ G values calculated using Rosetta (Fig.3).…”

RosettaDDGPrediction for high‐throughput mutational scans: From stability to binding

“…Using structural models to perform prediction of ΔΔ G s is a tantalizing perspective because of intrinsic limitations in the availability of experimental structures. This has been shown to be reliable to a good extent—for instance, using homology models with Rosetta allowed to achieve similar performance when comparing predictions with experimental ΔΔ G s, as long as the sequence identity of the template to the target protein was at least of 40% (Valanciute et al, n.d. ) and results obtained using Rosetta are relatively robust to the use of models (Blaabjerg et al, 2022 ; Valanciute et al, n.d. ). The advent of AlphaFold has revolutionized molecular modeling and structural biology (Jumper et al, 2021 ), resulting in models of 3D structures of proteins with quality comparable to that achievable with experimental approaches and useful in the context of computational biology, including the prediction of changes of free energy (Akdel et al, 2022 ).…”