Transformer Model Predicts And Designs Biosynthetic Gene Clusters

Researchers (Kawano et al.) publish on February 23, 2026 a transformer-based framework that models protein domains as tokens to predict and design biosynthetic gene clusters, training on four datasets including 2,492 MIBiG-validated BGCs. Models ranked true domains first over 50% of the time, exceeded 70% classification accuracy for major classes, and experimental expression produced an unknown cyclooctatin derivative.