FAMUS: A Few-Shot Learning Framework for Large-Scale Protein Annotation.

FAMUS, a contrastive learning framework, surpasses KEGG's KofamScan and InterProScan for protein annotation on sparse, realistic datasets.

• Why it matters: Accurate gene function prediction is essential for genomic analysis, but current tools struggle with setting meaningful thresholds and handling incomplete data, limiting reliability.
• What they did: The authors developed FAMUS, which compares query sequences to profile Hidden Markov Models, encoding full similarity profiles into vectors, and incorporates negative examples during training across four databases.
• The result: FAMUS enables robust, large-scale functional annotation without user-defined thresholds, supporting integration into genomic pipelines and outperforming existing methods on challenging datasets.