Combining Structural Modeling with Ensemble Machine Learning to Accurately Predict Protein Fold Stability and Binding Affinity Effects upon Mutation

Abstract

Advances in sequencing have led to a rapid accumulation of mutations, some of which are associated with diseases. However, to draw mechanistic conclusions, a biochemical understanding of these mutations is necessary. For coding mutations, accurate prediction of significant changes in either the stability of proteins or their affinity to their binding partners is required. Traditional methods have used semi-empirical force fields, while newer methods employ machine learning of sequence and structural features. Here, we show how combining both of these approaches leads to a marked boost in accuracy. We introduce ELASPIC, a novel ensemble machine learning approach that is able to predict stability effects upon mutation in both, domain cores and domain-domain interfaces. We combine semi-empirical energy terms, sequence conservation, and a wide variety of molecular details with a Stochastic Gradient Boosting of Decision Trees (SGB-DT) algorithm. The accuracy of our predictions surpasses existing methods by a considerable margin, achieving correlation coefficients of 0.77 for stability, and 0.75 for affinity predictions. Notably, we integrated homology modeling to enable proteome-wide prediction and show that accurate prediction on modeled structures is possible. Lastly, ELASPIC showed significant differences between various types of disease-associated mutations, as well as between disease and common neutral mutations. Unlike pure sequence-based prediction methods that try to predict phenotypic effects of mutations, our predictions unravel the molecular details governing the protein instability, and help us better understand the molecular causes of diseases.

Structural and physico-chemical effects of disease and non-disease nsSNPs on proteins

“…In line with experimental observations, it was found that naturally occurring mutations decrease binding affinity of spermine synthase homo-domain causing Snyder-Robinson syndrome [23], while enhancement of the binding between CLIC2 protein and ryanodine receptor causing an X-linked channelopathy [15, 18]. Genomic-scale investigations were also carried out with combined efforts of machine learning and statistical potentials [55]. Molecular dynamics studies, being computationally expensive, were applied to study specific cases only [56], including disruption of salt bridges between protein and substrate [57].…”

Integration of structural dynamics and molecular evolution via protein interaction networks: a new era in genomic medicine

“…Free energy differences upon binding calculated via thermodynamic integration and free energy perturbation approaches combined with molecular dynamics simulations are computationally expensive, particularly for large-scale protein complexes [41]. Therefore, many have developed in silico tools as a fast alternative to estimate ΔΔG using statistical energy functions based on known protein structures [42–44] and/or coupling with machine learning tools using training sets [45–47]. However, these calculations can be rather inaccurate, because local structural changes upon mutations are generally neglected [48,49].…”

Homology modeling and structure-based design improve hydrophobic interaction chromatography behavior of integrin binding antibodies

“…Ideally, such datasets would include negative data in order to robustly train predictive models and advance the field in the direction of data driven computational predictions, e.g., available protein thermostability benchmark datasets have allowed machine learning to be applied, resulting in accurate thermostability predictions. 18 Fortunately, data from multiple experiments for developability on 137 diverse mAbs in clinical trials or approved at the time of publication have been made available, 2 and a series of 31 adnectin variants with percent inclusion body data has recently been published. 19 …”