GeneEmbedLLM: Harnessing Large Language Models for Gene Functional Embedding and Association Prediction

GeneEmbedLLM generates universal, semantically-aware gene embeddings using an LLM that integrates multi-source gene annotations from five databases.

• Why it matters: Gene function is described in text across multiple sources, but existing representations often use single-source, non-textual data and miss that semantic information.
• What they did: The framework uses Longformer-based attribute-aware attention with contrastive learning to integrate structured text from five authoritative databases into gene embeddings.
• The result: Across downstream tasks spanning molecular interactions, phenotypic dependencies, and epigenetic regulation, it achieves competitive performance on all evaluated benchmarks and supports embedding quality via visualization, similarity, ablations, and parameter analysis.

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