iSEE: Interface structure, evolution, and energy‐based machine learning predictor of binding affinity changes upon mutations

Abstract

Quantitative evaluation of binding affinity changes upon mutations is crucial for protein engineering and drug design. Machine learning‐based methods are gaining increasing momentum in this field. Due to the limited number of experimental data, using a small number of sensitive predictive features is vital to the generalization and robustness of such machine learning methods. Here we introduce a fast and reliable predictor of binding affinity changes upon single point mutation, based on a random forest approach. Our method, iSEE, uses a limited number of i nterface S tructure, E volution, and E nergy‐based features for the prediction. iSEE achieves, using only 31 features, a high prediction performance with a Pearson correlation coefficient (PCC) of 0.80 and a root mean square error of 1.41 kcal/mol on a diverse training dataset consisting of 1102 mutations in 57 protein‐protein complexes. It competes with existing state‐of‐the‐art methods on two blind test datasets. Predictions for a new dataset of 487 mutations in 56 protein complexes from the recently published SKEMPI 2.0 database reveals that none of the current methods perform well (PCC < 0.42), although their combination does improve the predictions. Feature analysis for iSEE underlines the significance of evolutionary conservations for quantitative prediction of mutation effects. As an application example, we perform a full mutation scanning of the interface residues in the MDM2–p53 complex.

SAAMBE-3D: Predicting Effect of Mutations on Protein–Protein Interactions

“…First, we used the MDM2-p53 blind datasets presented in Table S8 in the supplementary information of reference [48]. The dataset consists of 33 mutations among which 7 exceeded the experimental detection limit and were removed.…”

“…We compared PCC of experimental and predicted ΔΔG for available methods. The methods considered for comparison here are iSee [48], mcSM [33], BindProfX [47], FoldX [41], mCSM-PPI2 [49] and MutaBind2 [50]. The PCC values for this dataset using iSee, mCSM, BindProfX and FoldX are taken from Geng et al's paper [48].…”

“…The last blind test set was the s487 dataset compiled by Geng et al [48] The dataset contains 487 mutations from 56 complexes. Geng et al [48] compared the performance of five different ∆∆G predictors (iSee, BindProfX, FoldX, mCSM and MutaBind2) on these 487 mutations. To compare SAAMBE-3D which was trained on SKEMPI v2.0 database, which in turn includes these mutations, we removed these mutations from our compiled dataset (dataset-1) and retrained our model on 80% of the remaining mutations.…”

“…In recent years, several methods [35,48,49] using machine learning were developed resulting in improved PCC ranging from 0.68 to 0.83. The advantage of machine learning (ML)-based methods is that it can utilize a variety of features based on structure, evolution, energy and others to predict ∆∆G, thus allowing for more extensive "parametrization".…”

“…mCSM method was reported to achieve a high PCC of 0.80 on 2317 single mutations from SKEMPI 1.1 database [30]. iSEE [48] is another predictor based on 31 features involving structure interface profile, energy and evolution-based features. It uses a position-specific scoring matrix generated from multiple sequence alignment to describe the evolutionary conservation of the mutation site.…”