Detecting remote homolog using structure alignment algorithms and machine learning

Remote homolog detection is a classic problem in Bioinformatics. It attempts to identify distantly related proteins sharing a similar structure. Methods that can accurately detect remote homologs benefit protein functional annotation. Recent computational advances in methods predicting the three-dimensional structure of a protein from amino acid sequences allow the massive use of structural data to develop new tools for identifying remote homologs. In this work, we created a discriminative SVM-based method based on structural alignment algorithms (FATCAT, TM-Align, and LovoAlign) to detect whether a protein is a remote homolog with any proteins in the SCOPe database. The final model showed a ROC AUC of 0.9191.