DDMut: predicting effects of mutations on protein stability using deep learning

Abstract

Understanding the effects of mutations on protein stability is crucial for variant interpretation and prioritisation, protein engineering, and biotechnology. Despite significant efforts, community assessments of predictive tools have highlighted ongoing limitations, including computational time, low predictive power, and biased predictions towards destabilising mutations. To fill this gap, we developed DDMut, a fast and accurate siamese network to predict changes in Gibbs Free Energy upon single and multiple point mutations, leveraging both forward and hypothetical reverse mutations to account for model anti-symmetry. Deep learning models were built by integrating graph-based representations of the localised 3D environment, with convolutional layers and transformer encoders. This combination better captured the distance patterns between atoms by extracting both short-range and long-range interactions. DDMut achieved Pearson's correlations of up to 0.70 (RMSE: 1.37 kcal/mol) on single point mutations, and 0.70 (RMSE: 1.84 kcal/mol) on double/triple mutants, outperforming most available methods across non-redundant blind test sets. Importantly, DDMut was highly scalable and demonstrated anti-symmetric performance on both destabilising and stabilising mutations. We believe DDMut will be a useful platform to better understand the functional consequences of mutations, and guide rational protein engineering. DDMut is freely available as a web server and API at https://biosig.lab.uq.edu.au/ddmut.

ProSTAGE: Predicting Effects of Mutations on Protein Stability by Using Protein Embeddings and Graph Convolutional Networks

“…Due to high-quality data is valuable for the robustness of the model, the previous test set has been incorporated into the training set to enhance the generalization ability of the model in recent methods . This makes it difficult to compare the different methods fairly.…”

AI-predicted protein deformation encodes energy landscape perturbation

“…Strain correlates with ∆∆G almost as well as tailored ∆∆G predictors.-To put the strain-stability correlations in context, we compare them with two state-of-the-art ∆∆G predictors, DDMut and FoldX (see Methods). 14,15 FoldX predicts ∆G from structure using empirical energy-based potentials; it is used to calculate ∆∆G by first a generating structure for the mutant based on a reference WT structure, which enables calculation of ∆G for both WT and mutant structures. DDMut uses a neural network to predict ∆∆G, using a reference structure and a mutation as input.…”

“…∆∆G predictors-We use two state-of-the-art methods to predict ∆∆G: FoldX 5, 21 and DDMut. 14 We use the API for the DDMut web server to predict ∆∆G. For FoldX, we use the BuildModel command to generate structures of the WT and mutant sequences and ∆∆G predictions.…”

Joint genotypic and phenotypic outcome modeling improves base editing variant effect quantification

“…To gain insights into unannotated variant impact, focusing on the LDLR class B repeats, we used the full wild-type LDLR structure from the AlphaFold Protein Structure Database 76,77 and the MODELLER 78 -generated mutant structures to calculate changes in interatomic interactions using Arpeggio 79 . Additionally, we predicted the effects of variants on protein stability (ΔΔ𝐺, negative value indicates destabilization) with DDMut 80 ( Supplementary Table 8 ). We found that the 26 significant LDLR class B variants induce more destabilizing effects, disrupt more hydrophobic interactions, have lower relative solvent accessibility 81 (0.041 of maximum residue solvent accessibility), and have higher wild-type residue depth as compared to the other observed variants in this region (Fig.…”