Predicting protein stability and solubility changes upon mutations: data perspective

Abstract

Understanding mutational effects on protein stability and solubility is of particular importance for creating industrially relevant biocatalysts, resolving mechanisms of many human diseases, and producing efficient biopharmaceuticals, to name a few. For in silico predictions, the complexity of the underlying processes and increasing computational capabilities favor the use of machine learning. However, this approach requires sufficient training data of reasonable quality for making precise predictions. This minireview aims to summarize and scrutinize available mutational datasets commonly used for training predictors. We analyze their structure and discuss the possible directions of improvement in terms of data size, quality, and availability. We also present perspectives on the development of mutational data for accelerating the design of efficient predictors, introducing two new manually curated databases FireProtDB and SoluProtMutDB for protein stability and solubility, respectively.