Sequence-Based Prediction for Protein Solvent Accessibility

When globular proteins fold into their characteristic three-dimensional structures, some amino acids are located on the surface, while others are situated in the protein core, where they cannot interact with molecules in the environment. Predicting the degree of solubility of amino acids provides insight into the function and relevance of residues. Residue accessibility is crucial for several protein functions, including enzymatic activity, allostery, multimer formation, binding to other molecules, and immunogenicity. We developed a novel sequence-based predictor for amino acid accessibility with features derived from three-dimensional protein structures. Several machine learning algorithms were tested, and the long short-term memory (LSTM) deep learning method demonstrated the best performance; thus, it was utilized to develop the freely available SolAcc tool. It showed superior performance compared to state-of-the-art predictors in a blind test.