Bacterial protein function prediction via multimodal deep learning • The study introduces DeepEST, a multimodal deep learning framework designed to predict bacterial protein functions by integrating protein structures, gene expression, and genomic location data, achieving significantly higher accuracy than traditional methods. • DeepEST combines a structure-based module, adapted from DeepFRI, with an expression-location module that captures bacterial-specific gene expression patterns, providing a more detailed functional annotation for bacterial proteins. • The framework is especially effective for predicting functions in hypothetical proteins, which lack experimental annotations, across 25 bacterial pathogens, highlighting its utility in advancing pathogen functional genomics. • Evaluated on diverse bacterial species, DeepEST outperformed state-of-the-art tools like BLAST, DIAMOND, and DeepGOplus, showing its superiority in handling the unique functional redundancy in bacterial genomes. • DeepEST’s structure-based data splitting strategy ensures robustness against data leakage, boosting its predictive power even when trained on diverse protein structures from various species. @kavican @kmborgwardt @AdamerMichael 💻Code: github.com/BorgwardtLab/Deep… 📜Paper: biorxiv.org/content/10.1101/… #ProteinFunction #Bioinformatics #BacterialGenomics #DeepLearning #ComputationalBiology