MIND-S, an AI tool for elucidating protein post-translational modifications in human diseases

Post-translational modifications (PTMs) serve as key regulatory mechanisms in many cellular processes; altered PTMs contribute significantly to disease pathogenesis in humans. Given the complexity and volume of PTM studies, computational methods for PTM prediction have been appreciated to be effective approaches to study PTMs. Here, we present MIND-S, an Artificial-Intelligence (AI) based PTM prediction tool utilizing protein-level information including sequence and structure. At its core, MIND-S employs a multi-head attention network and graph neural network (GNN); it assembles a 15-fold ensemble model in a multi-label strategy for PTM prediction. Trained on approximately 50,000 proteins and 26 PTM types, MIND-S demonstrates excellent functionality on simultaneous prediction of multiple types of PTM in a proteome setting while maintaining computational efficiency, benchmarking MIND-S documented its utility and performance. MIND-S also features an interpretation module, which assesses the relevance of each amino acid in making predictions and is able to capture known PTM motifs without any supervision. Furthermore, MIND-S enables examination of mutation effects on PTMs. Integrating with SNP information identified from GWAS, MIND-S has a utility to decode molecular insights and evaluate the impact of SNP from a PTM perspective. To promote transparency and reproducibility in AI research in PTM, we have made MIND-S and the dataset required for its training publicly available on GitHub. Additionally, we offer a detailed protocol with step-by-step instructions supporting users to navigate and implement MIND-S effectively. This holistic approach ensures FAIR of MIND-S, rendering a resource accessible and usable for the broader scientific community.