←
RxnLab
Single-step retrosynthesis
Beta
Enter your target molecule below
Help shape RxnLab:
share your feedback
(2 min, anonymous).
×
Target Product
(SMILES, name, InChI, InChIKey, or CAS; resolved via PubChem)
Examples:
Chloro-bipyridine nitrile
aspirin
caffeine
CAS 50-78-2
Paracetamol
Model
?
Which single-step retrosynthesis model runs the prediction. Each model has its own architecture and tunable parameters; switching models updates the controls below.
DiffAlign (align-absorbing)
LocalRetro
R-SMILES (RootAligned)
MEGAN
Number of Precursors
?
How many precursor sets the model samples. The raw samples are then deduplicated, and any set that can't be parsed by RDKit is discarded, so you'll usually see fewer cards than you requested.
Diffusion steps
?
How many denoising steps the model takes to produce each sample. More steps usually mean cleaner, more confident predictions but slower inference.
1
2
5
10
25
50
SMILES augmentations
?
How many randomized SMILES of the target the model translates and votes over. More augmentations give better, more stable predictions but slower inference.
Predict Precursors
Clear