Retrosynthetic analysis tools can be grouped into traditional rule/database systems, AI/deep-learning models, LLM-driven agents, open-source platforms, and other commercial tools. Below is a synthesized compilation.
5 Replies
AdministratorSite Administrator
1. Traditional Classic Computational Tools (Rule- and Database-Based)
These rely on curated reaction databases, expert-encoded rules, and literature/patent data for known routes and conditions.
ToolDeveloper/SourceCore FeaturesPrimary UsesReaxysElsevierBuilt-in retrosynthesis module; massive literature + patent dataKnown-compound synthesis lookup; experimental step tracingSciFinder-nCAS (ACS)Retrosynthesis on the most comprehensive CAS reaction databaseLiterature route search; condition verification; supplier queriesSynthia (formerly Chematica)Merck/Sigma-Aldrich>100,000 hand-coded reaction rules; logical, highly interpretableComplex-molecule route design; industrial R&D; route optimizationLHASAAcademic/teachingClassic expert system using “strategic bond disconnection” rulesTeaching demonstrations; academic research
AdministratorSite Administrator
2. AI / Deep-Learning Models (Single-Step & Multi-Step Planning)
These treat retrosynthesis as a sequence-to-sequence or graph-based prediction task trained on large reaction datasets.
Synthon (Insilico Medicine)
2.6-billion-parameter LFM using root-aligned SMILES strategy; Top-1 accuracy 66.23–74.12 %.
Main use: high-accuracy single-step retrosynthesis prediction.
C3LM (academic)
Chemistry-reasonableness-aware LLM trained on 45.6 million reactions with ChemCensor reward mechanism.
Main uses: handling out-of-distribution data and complementing classical models.
Molecular Transformer (IBM Research)
Sequence-to-sequence translation task (product SMILES → reactant SMILES).
Main uses: open benchmark model and rapid single-step bond-breaking validation.
RxnFP (IBM Research)
Same series as Molecular Transformer.
Main uses: reaction fingerprint generation and retrosynthesis prediction.
G2Gs (Graph-to-Graph) (academic)
Graph neural network that predicts bond-breaking positions and generates reactants.
Main use: molecule-graph-level retrosynthetic analysis.
AdministratorSite Administrator
3. LLM-Driven Intelligent Agent Systems (Constraint-Aware & Automated)
These combine large language models with external tools, real-time constraint checking, and (in some cases) laboratory automation.
LARC (academic, frontier)
First “Agent-as-a-Judge” LLM framework that evaluates constraints in real time.
Main uses: avoiding toxic intermediates and hazardous reactions (72.9 % success rate, 3.6× faster).
ChemCrow (academic)
LLM connected to RDKit and other external chemistry software.
Main uses: design–make–test closed loops and shortening R&D cycles.
Organa (academic)
Combines LLM with laboratory automation equipment.
Main uses: automated experiment design and execution.
GIT-Mol (academic)
Multimodal model aligning SMILES, molecular graphs, 3-D coordinates, and natural language.
Main uses: retrosynthesis + property prediction + literature Q&A.
nach0 (academic)
Multimodal model in the same family as GIT-Mol.
Main use: comprehensive chemical information processing.
AdministratorSite Administrator
4. Open-Source / Freely Available Platforms
These support local deployment, customization, and academic use.
ASKCOS (MIT)
Based on Transformer and neural-network strategies; supports multi-step search.
Main uses: academic research, local deployment, single-step and multi-step retrosynthesis.
AIZynthFinder (AstraZeneca)
Monte-Carlo tree search (MCTS) that can be combined with any single-step prediction model.
Main uses: local deployment, secondary development, and multi-step route search.
IBM RXN (IBM)
Transformer model available after free registration.
Main uses: retrosynthesis prediction and forward reaction prediction.
SynthI (open community)
RDKit-based and locally deployable.
Main use: lightweight retrosynthetic analysis.
RDChiral (open community)
RDKit-based retrosynthetic template library.
Main use: underlying engine for custom development.
Chemy AI (third-party)
No API barriers or usage limits; integrates multi-step retrosynthesis engines.
Main uses: general workflows, structure parsing, and database scanning.
AdministratorSite Administrator
5. Other Commercial / Third-Party Tools
Spaya (Iktos)
AI-driven with emphasis on synthesizability scoring and route ranking.
Main uses: automated route generation and optimization.
Pistachio (NextMove Software)
Automatically extracts reaction data from patents and literature.
Main use: data foundation for building retrosynthesis engines.
ChemPlanner (Wiley)
Built on the Wiley reaction database.
Main use: multi-step route planning.
Chemical.AI (Chinese vendor)
AI synthesis planning.
Main use: industrial-scale synthetic route design.
Molecule.one (third-party)
AI synthesis planning.
Main use: large-scale route discovery.