The influence of reduced amino acid alphabets on prediction orthologous protein thermostability

Sequence features play a vital role in determining protein thermostability. Considering the advantages of reduced amino acid alphabets (RAAs) in reducing data complexity and retaining key sequence information, we evaluate the performance of 672 RAAs on prediction orthologous protein thermostability. We calculate the Amino Acid Composition, Dipeptide Composition, and Tripeptide Composition of the reduced sequence features, and use the random forest model to make predictions. The results show that 10 RAAs, selected using fuzzy clustering, are effective in predicting thermostability differences between orthologous protein pairs, significantly improving prediction efficiency. Further, the melting temperature difference Delta Tm caused by point mutations is predicted, and it is found that the RAA of EQ-H-K-DN-IL-P-T-FY-M-R-S-W-A-C-G-V could fit the tiny thermostability change caused by point mutations. Our work showcases that the reduction methods based on fuzzy clustering can effectively retain the key sequence features that affect protein thermostability, resulting in reducing the computational complexity and increasing the prediction accuracy.