Predicting protein thermostability change upon mutation is crucial for understanding diseases and designing therapeutics. However, accurately estimating Gibbs free energy change of the protein remained a challenge. Some methods struggle to generalize on examples with no homology and produce uncalibrated predictions. Here we leverage advances in graph neural networks for protein feature extraction to tackle this structure–property prediction task. Our method, BayeStab, is then tested on four test datasets, including S669, S611, S350, and Myoglobin, showing high generalization and symmetry performance. Meanwhile, we apply concrete dropout enabled Bayesian neural networks to infer plausible models and estimate uncertainty. By decomposing the uncertainty into parts induced by data noise and model, we demonstrate that the probabilistic method allows insights into the inherent noise of the training datasets, which is closely relevant to the upper bound of the task. Finally, the BayeStab web server is created and can be found at: http://www.bayestab.com. The code for this work is available at: https://github.com/HongzhouTang/BayeStab.

MotivationPredicting protein stability change upon variation through computational approach is a valuable tool to unveil the mechanisms of mutation-induced drug failure and help to develop immunotherapy strategies. However, some machine learning based methods tend to be overfitting on the training data or show anti-symmetric biases between direct and reverse mutations. Moreover, this field requires the methods to fully exploit the limited experimental data.ResultsHere we pioneered a deep graph neural network based method for predicting protein stability change upon mutation. After mutant part data extraction, the model encoded the molecular structure-property relationships using message passing and incorporated raw atom coordinates to enable spatial insights into the molecular systems. We trained the model using the S2648 and S3412 datasets, and tested on the Ssym and Myoglobin datasets. Compared to existing methods, our proposed method showed competitive high performance in data generalization and bias suppression with ultra-low time consumption. Furthermore, method was applied to predict the Pyrazinamide’s Gibbs free energy change for a real case study.Availabilityhttps://github.com/shuyu-wang/ProS-GNN.Contactvincentwang622@126.com