Environment-Aware DNA Language Model for Stress-Responsive Genomic Prioritization in Maize

An environment-aware DNA-LLM (AgroNT) for maize identifies putative stress-responsive genomic regions by detecting stress-conditioned representation shifts and attention patterns across 4,000+ hybrids.

• Why it matters: Stress resilience is hard to map because genotype-by-environment effects complicate linking DNA regions to heat, drought, and combined stress responses.
• What they did: The framework adds stress-specific prompt tokens and uses parameter-efficient fine-tuning on AgroNT, then prioritizes regions using stress-induced embedding shifts versus control plus stress-token attention patterns on the G2F dataset.
• The result: Prioritized sequences show enriched stress-associated motifs and high-attention regions, and nearby genes align with transcriptional regulation, signaling, and metabolic pathways using spatiotemporal differential expression and stress-linked QTL evidence.

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