Closed h-midlothian closed 3 years ago
Retro* uses trained models for one-step retrosynthesis prediction, so you can use our codebase to do one-step predictions as well.
The problem is that the one-step interface is not exposed and you may need to write extra code for it to work. To do that, you need to first load the learned one-step model from disk using the one_step = prepare_mlp(mlp_templates, mlp_model_dump)
function, and call the model via result = one_step(product_smiles)
. The result you will obtain is a dictionary containing scores
, reactants
, templates
keys, through which you can access the probabilities, reactant smiles, and reaction templates of up to k candidates respectively.
Examples: https://github.com/binghong-ml/retro_star/blob/master/retro_star/api.py#L31 https://github.com/binghong-ml/retro_star/blob/master/retro_star/alg/molstar.py#L37
Retro* uses trained models for one-step retrosynthesis prediction, so you can use our codebase to do one-step predictions as well.
The problem is that the one-step interface is not exposed and you may need to write extra code for it to work. To do that, you need to first load the learned one-step model from disk using the
one_step = prepare_mlp(mlp_templates, mlp_model_dump)
function, and call the model viaresult = one_step(product_smiles)
. The result you will obtain is a dictionary containingscores
,reactants
,templates
keys, through which you can access the probabilities, reactant smiles, and reaction templates of up to k candidates respectively.Examples: https://github.com/binghong-ml/retro_star/blob/master/retro_star/api.py#L31 https://github.com/binghong-ml/retro_star/blob/master/retro_star/alg/molstar.py#L37
Solved my problem. Thanks a lot
Hi, thanks for your code and paper. I just wonder if I can run a retrosyn prediction for a molecule with only a one-step route as the output, instead of the default multi-step routes given in the example?