RDT v3.8.1
This release makes core algorithmic and engineering work developed internally at BioInception PVT LTD available to the research community, students, and the broader cheminformatics ecosystem under LGPL v3.
What is RDT?
Reaction Decoder Tool (RDT) is a deterministic, training-free reaction atom-atom mapping engine. It extracts bond changes and reaction centres from chemical reactions — no machine learning, no training data, fully reproducible.
Highlights: v3.x→ v3.8.1
Algorithmic
- Identity pre-filter: direct atom mapping for unchanged molecules, eliminating symmetry-induced spurious bond changes
- Stereo-correct identity detection via SMSD MolGraph canonical SMILES
- Pre-filter pipeline (identity → atom-count ratio → Tanimoto similarity) reducing MCS workload without compromising accuracy
- Formal algorithm description published (
ALGORITHM.md)
Engine & Dependencies
- SMSD upgraded to 6.9.1
- Canonical SMILES for identity detection now uses SMSD
MolGraph— stereo-aware and consistent with internal MCS canonicalisation - Java 21 full compatibility; CDK 2.12
- CI build validated (Java 21 Temurin)
Benchmark — Golden Dataset (Lin et al. 2022, 1,851 reactions)
| Metric | Score |
|---|---|
| Chemically-equivalent atom mapping | 99.2% |
| Mol-map exact | 76.8% |
| Bond-change exact | 99.2% |
| True chemistry misses | 0.8% |
RDT leads all published deterministic tools on chemically-equivalent accuracy (RDTool 76.18%, ChemAxon 70.45%).
Installation
# Download jar
https://github.com/asad/ReactionDecoder/releases/tag/v3.8.1
# One-line API
ReactionResult result = RDT.map("CC(=O)O.OCC>>CC(=O)OCC.O");Author: Dr. Syed Asad Rahman
Organisation: BioInception PVT LTD
Contact: asad.rahman@bioinceptionlabs.com
License: LGPL v3