RDT v3.7.0

Reaction Decoder Tool (RDT) v3.7.0

Deterministic atom-atom mapping engine. No training data. No machine learning.

Golden Dataset Benchmark (Lin et al. 2022, 1,851 reactions)

Tool	Exact Match	Atom Accuracy	Training Data	Deterministic
RDT v3.7.0	82.0%	96.4%	None	Yes
RXNMapper	83.74%	-	Unsupervised	No
RDTool (published, 2016)	76.18%	-	None	Yes
ChemAxon	70.45%	-	Proprietary	Yes

Reference: Lin A, Dyubankova N, Madzhidov TI, et al. Molecular Informatics 41(4):e2100138, 2022

What's New in v3.7.0

• Golden dataset benchmark — 1,851 manually curated reactions with combined metrics (exact match, atom accuracy, bond accuracy, quality score, parsimony)
• SMSD 6.7.0 — Optimized circular fingerprints (ECFP/FCFP), reaction-aware MCS, batch constrained MCS
• SMSD adapter optimization — Reuse single SMSD object for substructure + MCS fallback, eliminating redundant object creation
• Security hardening — XXE protection, path traversal defense, resource leak fixes
• Memory optimization — Fingerprint caching, ThreadSafeCache cleanup in finally blocks, thread-safe molecule cloning
• Reduced CDK footprint — Removed cdk-fingerprint dependency entirely, all fingerprinting via SMSD
• Algorithm documentation — Formal 9-stage pipeline description (ALGORITHM.md) suitable for publication

Performance

Metric	Value
Mapping success	100% (1,851/1,851 golden dataset)
Atom-level accuracy	96.4%
Mapping speed	~3.4 reactions/sec
Test suite	164 tests, 100% pass
RXN coverage	598/599 (99.8%) on internal test set

Dependencies

• SMSD 6.7.0 (com.bioinceptionlabs:smsd)
• CDK 2.12 (15 individual modules, lightweight)
• Java 21+

How to Cite

Rahman SA, Torrance G, Baldacci L, et al. "Reaction Decoder Tool (RDT): Extracting Features from Chemical Reactions." Bioinformatics 32(13):2065-2066, 2016. DOI: 10.1093/bioinformatics/btw096

License

GNU Lesser General Public License (LGPL) v3.0