AI Insight
RETROSPECT is a new computational system for retrosynthesis prediction that combines a Transformer-based proposal model (ChemAlign Transformer) with a LambdaMART reranking algorithm to suggest synthetic routes for target molecules. On the USPTO-50K benchmark dataset of 5,007 reactions, the system achieved 55.00% top-1 accuracy and 86.18% top-10 accuracy for proposing correct synthetic pathways, with the reranking model improving top-1 accuracy to 59.4% when selecting from pooled candidates. Feature analysis revealed that proposal confidence scores and reaction template frequency statistics were the most important factors for reranking, while density functional theory (DFT) calculations provided minimal additional benefit.
Why it matters
This work advances computer-aided synthesis planning in drug discovery and chemical manufacturing by improving the accuracy of automated retrosynthesis prediction. The modular architecture allows the system to be integrated into existing retrosynthesis platforms, potentially accelerating the design of synthetic routes for new pharmaceutical compounds and specialty chemicals.
arXiv:2606.07181v1 Announce Type: cross
Abstract: Single-step retrosynthesis needs both accurate first-ranked suggestions and candidate lists that are rich enough for downstream selection. We study this as a proposal-selection decomposition. Our system, RETROSPECT, combines a single Transformer proposal model, which we call the ChemAlign Transformer, with a LambdaMART reranker over structural, reaction-template, upstream-score, and optional DFT-derived descriptors. The generator is trained with hybrid root-aligned and random SMILES augmentation, Pre-LayerNorm, tied embeddings, exponential moving average weights, and a differentiable atom-balance auxiliary loss. On the full USPTO-50K test set of 5,007 reactions, the generator reaches 55.00% top-1 and 86.18% top-10 exact-match accuracy with 99.86% top-1 validity. On the merged candidate-pool benchmark used for reranking, which contains 5,007 test products and about 111 candidates per product, a LambdaMART model trained on the structural feature set reaches 59.4% top-1 with 0.7171 mean reciprocal rank. Feature ablations show that upstream proposal score and template-frequency statistics provide most of the reranking signal, while DFT and reaction-center DFT features provide smaller and less consistent gains. These results support a modular view of retrosynthesis: stronger single-model proposal and learned candidate selection are complementary, and the proposal model can serve as a drop-in component for ensemble systems such as RetroChimera (Maziarz et al., 2024)
Source: RETROSPECT: RETROsynthesis via Sequential Prediction, and Chemically Transformed-ranking