Protein large language model-assisted one-to-one gene homology mapping in cross-species single-cell transcriptome integration.

Protein large language model-based gene homology mapping increases homology gene pairs by approximately 8% while maintaining one-to-one mappings in cross-species single-cell transcriptome integration.

• Why it matters: Accurate gene correspondence is crucial for comparative analysis across species, but current methods often produce many-to-many mappings that can obscure cell-type identities and evolutionary signals, limiting biological insights.
• What they did: The authors developed a fused mapping strategy combining pLLM-derived representations with sequence similarity, enforcing one-to-one mappings and benchmarking across nine datasets, 11 species, and over 3.2 million cells.
• The result: This approach enhances homology detection, identifies new cell-type marker pairs, and provides a more accurate, interpretable framework for cross-species transcriptome analysis.