Chemistry & ADMET · Sagawa & Kaneko · IBM RXN
Reaction prediction and fingerprints
Predicts forward reactions and retrosynthetic routes for small molecules, and generates reaction fingerprints.
What Reaction prediction and fingerprints is
Reaction prediction and fingerprints (rxnfp) is a chemistry tool developed by Sagawa & Kaneko and IBM RXN for small-molecule chemistry. It provides forward reaction prediction, retrosynthetic planning, and RXNFP fingerprints designed to compare chemical reactions.
The model frames chemical reaction prediction as a translation problem: submitting reactants generates the corresponding product, whilst providing a target molecule generates potential precursors. By learning from published literature, it assists researchers in evaluating whether a computationally generated compound is plausibly makeable before allocating experimental laboratory resources.
How it works
Reaction prediction is formulated as a sequence translation task using SMILES (Simplified Molecular Input Line Entry System) strings. Given reactant structures alongside optional reagents and solvents, the model translates them into product structures, or conversely translates target products into candidate starting materials. The model is trained on chemical reactions extracted from United States Patent and Trademark Office (USPTO) patents.
Because a single chemical molecule often possesses multiple viable disconnection sites, retrosynthetic prediction returns a ranked set of plausible precursors rather than a single solution. Each generated output is checked to confirm that its SMILES string parses correctly.
Key facts
- Version
- 2.0
- Domain
- Chemistry
- Modalities
- Small molecule
- Tasks
- Reaction prediction, retrosynthesis
- Training data
- USPTO patent reactions
- GPU memory
- 8 GB
What you can use it for
- Forward reaction outcome prediction
- Forecast the likely chemical product given starting reactants and optional solvent or reagent SMILES strings.
- Retrosynthetic route generation
- Propose and rank candidate synthetic precursors and disconnections for a desired target molecule.
- Synthetic feasibility screening
- Assess whether computationally proposed molecules correspond to known literature patterns before ordering or preparing them in the laboratory.
- Chemical reaction comparison
- Compare distinct chemical transformations using RXNFP reaction fingerprints.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
Reaction prediction and fingerprints
Strengths and limits
Good at
- Returns a ranked set of retrosynthetic proposals to accommodate multiple plausible synthetic disconnections.
- Automatically verifies that all returned product and precursor candidate SMILES parse correctly.
- Handles both forward synthesis and retrosynthetic planning within a single unified translation framework.
Watch out for
- Predictions reflect patent literature precedent and do not constitute an experimental guarantee that a proposed reaction step will succeed.
- Input molecules, reagents, and solvents must be provided in text-based SMILES format.
Inputs and outputs
You provide
Reaction or molecule (SMILES)
Reagents and solvent (SMILES)optional
Optional but strongly recommended for forward prediction, e.g. C1CCOC1 for THF.
What to computeoptional
Compare withoptional
A second reaction. With fingerprints selected, the answer is how alike the two are.
Predictionsoptional
You get back
Ranked SMILES with confidence, each checked for whether it parses.
Licence
Commercial use permitted · MIT
The MIT licence allows use in commercial research and products.
Questions
Can I use Reaction prediction and fingerprints commercially?
Commercial use permitted. The MIT licence allows use in commercial research and products.
How do I run Reaction prediction and fingerprints without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on 1× A100 16GB, typically in 2.0s, and a run costs about 6 credits. Visitors without an account get 1,000 credits a month.
What training data does the model use?
The model is trained on reaction records extracted from USPTO patent literature.
What formats are used for model inputs and outputs?
Inputs are supplied as text-based SMILES strings for molecules, reactants, and reagents. Outputs are returned as a table of ranked candidate SMILES strings with associated confidence values.
Does a successful prediction guarantee the reaction will work in the laboratory?
No. The model predicts what the patent literature would suggest and indicates whether a molecule is plausibly makeable, but it does not claim that any particular reaction step will work experimentally.
How are retrosynthetic suggestions evaluated?
The model provides a ranked list of candidate precursor sets alongside confidence scores, and each suggested output is parsed to confirm validity.
Written from Reaction prediction and fingerprints's own documentation. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.