Potentially novel computational question (novelty not claimed globally):

Across five grass OsNAC25 homologs from rice, foxtail millet, sorghum, finger millet, and teff, does divergence in the C-terminal regulatory region produce a stronger protein-language-model separation than the conserved N-terminal NAC DNA-binding core, while hosted structure prediction preserves a shared high-confidence core and lower-confidence lineage-specific tails?

This is a sequence-, embedding-, alignment-, domain-, and structure-based plant genomics study. It is not a literature report. Model outputs are hypothesis-tier predictions, not experimental observations.
