RETROSPECT: RETROsynthesis via Sequential Prediction, and Chemically Transformed-ranking
Researchers have developed RETROSPECT, a novel system for chemical retrosynthesis that improves prediction accuracy and candidate selection. The system utilizes a Transformer-based proposal model, ChemAlign, combined with a LambdaMART reranker. This approach achieved 55.00% top-1 exact-match accuracy on the USPTO-50K dataset, demonstrating a significant advancement in predicting chemical reactions. AI
IMPACT Enhances AI's capability in scientific discovery, potentially accelerating drug development and chemical research.