Rapid protein stability prediction using deep learning representations

Abstract

Predicting the thermodynamic stability of proteins is a common and widely used step in protein engineering, and when elucidating the molecular mechanisms behind evolution and disease. Here, we present RaSP, a method for making rapid and accurate predictions of changes in protein stability by leveraging deep learning representations. RaSP performs on-par with biophysics-based methods and enables saturation mutagenesis stability predictions in less than a second per residue. We use RaSP to calculate ≈8.8 million stability changes for nearly all single amino acid changes in 1,381 human proteins, and examine variants observed in the human population. We find that variants that are common in the population are substantially depleted for severe destabilization, and that there are substantial differences between benign and pathogenic variants, highlighting the role of protein stability in genetic diseases. RaSP is freely available and enables large-scale analyses of stability in experimental and predicted protein structures.

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

“…In the case of MSH2 and MLH1, for example, it has been shown that a predicted destabilization of more than 3 kcal/mol is sufficient to cause cellular degradation of the proteins (Abildgaard et al, 2019 ; Nielsen et al, 2017 ). Similar observations have been recently done in another recent work on different proteins with benign variants featuring predicted changes in stability in the range of 0.9–2.7 kcal/mol (Blaabjerg et al, 2022 ).…”

“…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 ).…”

“…We performed the same analyses on the dataset obtained using the cartesian2020 protocol, which showed similar trends overall (Figure S1 ). Additionally, for comparison, we used a recent deep learning tool, which aims at simulating the cartddg protocol, that is, RaSP (Blaabjerg et al, 2022 ) (Figure S3 ). Here, we also noticed a remarkable performance both compared to both the experimental values and the Rosetta predictions.…”

Learning the shape of protein micro-environments with a holographic convolutional neural network

“…The H-CNN learned representations of amino acid neighborhoods could be used as input to a supervised algorithm to learn a more accurate model for mutational effects in proteins; a similar approach has been used to model the stability effect of mutations in ref. [65]. Moreover, the all-atom representation of protein structures used to train H-CNN allows for generalizability, e.g.…”

Deep Local Analysis deconstructs protein - protein interfaces and accurately estimates binding affinity changes upon mutation

“…They exhibit exciting potential for a broad range of protein-related problems [40, 62, 43, 44, 56, 78]. Beyond sequence information, self-supervised learning-based approaches have leveraged the protein and protein complex 3D structures available in the Protein Data Bank (PDB) [5] for fixed-backbone protein design [2, 28, 11], for predicting protein stability [6, 77], and for assessing the impact of mutations on protein-protein interactions [42]. In particular, in [42], a graph neural network is trained to reconstruct disturbed wild-type and mutated complex structures represented as graphs.…”