DeepDDG: Predicting the Stability Change of Protein Point Mutations Using Neural Networks

Accurately predicting changes in
protein stability due to mutations
is important for protein engineering and for understanding the functional
consequences of missense mutations in proteins. We have developed
DeepDDG, a neural network-based method, for use in the prediction
of changes in the stability of proteins due to point mutations. The
neural network was trained on more than 5700 manually curated experimental
data points and was able to obtain a Pearson correlation coefficient
of 0.48–0.56 for three independent test sets, which outperformed
11 other methods. Detailed analysis of the input features shows that
the solvent accessible surface area of the mutated residue is the
most important feature, which suggests that the buried hydrophobic
area is the major determinant of protein stability. We expect this
method to be useful for large-scale design and engineering of protein
stability. The neural network is freely available to academic users
at http://protein.org.cn/ddg.html.

精准预测突变引发的蛋白质稳定性变化，对于蛋白质工程研究以及解析蛋白质错义突变的功能影响具有重要意义。本研究开发了基于神经网络的方法DeepDDG，用于预测单点突变引发的蛋白质稳定性变化。该神经网络基于超过5700条经人工整理的实验数据点进行训练，在三个独立测试集上的皮尔逊相关系数（Pearson correlation coefficient）可达0.48至0.56，性能优于其余11种方法。对输入特征的详细分析表明，突变残基的溶剂可及表面积是最为关键的输入特征，这提示埋藏疏水区域是决定蛋白质稳定性的核心因素。本研究预期该方法可应用于大规模蛋白质稳定性设计与工程改造，且该神经网络已向学术用户免费开放，访问地址为http://protein.org.cn/ddg.html。