rxnutils.routes.deepset package

Submodules

rxnutils.routes.deepset.featurizers module

Module with featurizers for molecules and reactions

rxnutils.routes.deepset.featurizers.collect_reaction_features(reaction_data, target_fingerprint, reaction_class_ranks, featurizer, prefactor=1.1, default_rank=5, default_cls='0.0')

Collect all reaction features and compute the prior reaction feasibility score

Params reaction_data:

the reactions of the route

Params target_fingerprint:

the precomputed fingerprint of the target

Params reaction_class_ranks:

the ranks of the reaction classes

Params featurizer:

the featurizer to use for reactions

Params prefactor:

the factor for the feasibility score

Params default_rank:

the rank of reactions not parameterized

Params default_cls:

the class of reactions not parameterized

Returns:

the prior feasibility score

Returns:

the reaction features

Parameters:

  • reaction_data (List[Dict[str, Any]])

  • target_fingerprint (ndarray)

  • reaction_class_ranks (Dict[str, int])

  • featurizer (Callable[[Dict[str, Any]], ndarray])

  • prefactor (float)

  • default_rank (int)

  • default_cls (str)

Return type:

Tuple[float, ndarray]

rxnutils.routes.deepset.featurizers.default_reaction_featurizer(reaction_data, radius=2, numbits=64)

Given an item of reaction data, returned by the reaction_data method of SynthesisRoute this features the reaction using a difference ECFP

Currently it featurizes the reaction in the retro-sense

Params reaction_data:

the reaction as a dictionary

Params radius:

the radius of the fingerprint

Params numbits:

the number of bits

Returns:

the fingerprint

Parameters:

  • reaction_data (Dict[str, Any])

  • radius (int)

  • numbits (int)

Return type:

ndarray

rxnutils.routes.deepset.featurizers.ecfp_fingerprint(mol, radius=2, numbits=64)

Computes a ECFP with a given radius and length

Params mol:

the RDKit molecule

Params radius:

the radius of the fingerprint

Params numbits:

the number of bits

Returns:

the fingerprint

Parameters:

  • mol (Mol)

  • radius (int)

  • numbits (int)

Return type:

ndarray

rxnutils.routes.deepset.featurizers.reaction_difference_fingerprint(reaction, radius=2, numbits=64)

Computes the difference fingerprint of given reaction

Params reaction:

the reaction to featurize

Params radius:

the radius of the fingerprint

Params numbits:

the number of bits

Returns:

the fingerprint

Parameters:
Return type:

ndarray

rxnutils.routes.deepset.scoring module

Module containing routines for Deepset route scoring

This is an implementation of the code used in the following publication, original implementation by Guo Yujia

Yujia G, Kabeshov M, Le THD, Genheden S, Bergonzini G, Engkvist O, et al. A Deep Learning with Expert Augmentation Approach for Route Scoring in Organic Synthesis. ChemRxiv. 2024; doi:10.26434/chemrxiv-2024-tp7rh

class rxnutils.routes.deepset.scoring.DeepsetModelClient(model_path)

Bases: object

Interface for an in-memory instance of a Deepset model for route scoring

Params model_path:

the path to an ONNX model file

Parameters:

model_path (str)

rxnutils.routes.deepset.scoring.deepset_route_score(route, model_client, scscorer, reaction_class_ranks)

Scores a synthesis route using a Deepset model

Currently, it uses defaults for featurizers as described in the original publication.

Params route:

the route to score

Params model_client:

the interface to the model

Params scscorer:

the interface to the SCScore

Params reaction_class_ranks:

the lookup from reaction class to rank

Returns:

the computed score

Parameters:
Return type:

float

Module contents