Reaction Prediction & Synthesis

IBM RXN team's overview of the Molecular Transformer approach to reaction SMILES prediction and retrosynthesis.

Comprehensive review of CASP approaches: rule-based, template-based, and ML-based methods from 2021.

Free IBM web platform for forward prediction, retrosynthesis, and procedure extraction using transformer models. No coding required.

Interactive tutorial for MIT's synthesis planning platform: one-step retrosynthesis, multi-step tree search, and condition recommendation.

Applies the transformer to reaction SMILES for forward prediction and retrosynthesis, achieving state-of-the-art top-1 accuracy.

Integrates ML retrosynthesis with robotic synthesis execution — a precursor to self-driving labs.

High-throughput Suzuki coupling dataset used in Tutorial 11 for yield prediction and Bayesian optimisation.

Analysis of reaction datasets: USPTO, Reaxys, and proprietary pharma data for synthesis planning.

Free transformer-based platform for forward reaction prediction, retrosynthesis, and procedure extraction.

Open-source CASP platform with one-step retrosynthesis, multi-step tree search, and condition recommendation.

Differential Reaction Fingerprints for reaction classification and yield prediction. Fast drop-in for Morgan fingerprints in reaction tasks.

RDKit's reaction SMARTS engine for template matching, reaction enumeration, and product prediction.

50,000 atom-mapped reactions from USPTO patents, divided into 10 reaction types. The standard benchmark for one-step retrosynthesis and forward prediction models.

Commercial reaction database (Elsevier) with 50M+ reactions from the primary literature. Widely used in industry for reaction prediction training data (subscription required).

Community-curated open database of chemical reactions with structured data for conditions, yields, and provenance. Growing free alternative to Reaxys.