Identifying Protein-Nucleotide\nBinding Residues via\nGrouped Multi-task Learning and Pre-trained Protein Language Models

The accurate identification of protein-nucleotide binding\nresidues\nis crucial for protein function annotation and drug discovery. Numerous\ncomputational methods have been proposed to predict these binding\nresidues, achieving remarkable performance. However, due to the limited\navailability and high variability of nucleotides, predicting binding\nresidues for diverse nucleotides remains a significant challenge.\nTo address these, we propose NucGMTL, a new grouped deep multi-task\nlearning approach designed for predicting binding residues of all\nobserved nucleotides in the BioLiP database. NucGMTL leverages pre-trained\nprotein language models to generate robust sequence embedding and\nincorporates multi-scale learning along with scale-based self-attention\nmechanisms to capture a broader range of feature dependencies. To\neffectively harness the shared binding patterns across various nucleotides,\ndeep multi-task learning is utilized to distill common representations,\ntaking advantage of auxiliary information from similar nucleotides\nselected based on task grouping. Performance evaluation on benchmark\ndata sets shows that NucGMTL achieves an average area under the Precision-Recall\ncurve (AUPRC) of 0.594, surpassing other state-of-the-art methods.\nFurther analyses highlight that the predominant advantage of NucGMTL\ncan be reflected by its effective integration of grouped multi-task\nlearning and pre-trained protein language models. The data set and\nsource code are freely accessible at: https://github.com/jerry1984Y/NucGMTL.