Bacterial protein function prediction via multimodal deep learning.

DeepEST, a multimodal bacterial protein function predictor, achieves a protein-centric F max of 0.689 and term-centric micro-AUPRC of 0.756 on a 25-species benchmark.

• Why it matters: Many bacterial proteins lack functional annotation, limiting understanding of survival under diverse stresses. Existing predictors underuse bacterial-specific signals from expression and genomic context.
• What they did: DeepEST combines a gene-expression/location module with a structure-and-sequence module, then fuses them via a learnable weighted linear combination and a masked loss to fine-tune structure-based learning for bacterial species. It is evaluated on a 25-species benchmark using only amino acid sequence or protein structure.
• The result: DeepEST outperforms prior methods by 4.4% (F max) and 11.6% (micro-AUPRC) versus DeepFRI, and it predicts GO terms for unclassified hypothetical proteins across 25 human bacterial pathogens.

AI summary of the abstract · may contain errors · not medical advice · read the paper.