Domain-Aware Enzyme Classification via Multiple Instance Learning.

MIL-BC, a domain-bag multiple instance learning framework for EC prediction, improves F1 by 0.03–0.17 over a sequence-based baseline across four test sets.

• Why it matters: Whole-sequence embeddings can blur local functional signals, and structure-based EC methods are limited by dependence on accurate 3D structures.
• What they did: The approach uses DCTdomain to automatically generate domain instances, aggregates domain evidence into protein-level EC predictions, and learns interpretable domain weights; MIL-BC+ also combines domain and global embeddings.
• The result: Domain-centric modeling yields the largest gains for proteins with multiple enzymatic functions, enabling scalable, interpretable EC prediction without structural data or curated domain labels.

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