ChemFeat and MolPred¶

Python packages for developing predictors of molecular properties.

Links¶

Author¶

Jan-Michael (Mike) Rye

Inria Research Engineer, SED-LYS, Lyon.
Permanent member of AIstroSight.
Creator of ChemFeat & MolPred.

ChemFeat¶

A simple Python package and command-line tool for generating feature vectors from molecules

Usage Overview¶

Provide a list of molecules as InChi strings.
Select and configure feature sets using a simple YAML configuration file
Use ChemFeat to generate a pandas dataframe or CSV file with the calculated features.

Notable Features¶

Already supports ~70 feature sets thanks to RDKit and PaDEL Descriptor.
Modular design facilitates addition of new feature sets.
Calculated features are cached in a database to avoid redundant calculations when rerunning code.

Command-Line Tool¶

Documentation

Example¶

# features.yaml

# QED feature calculator.
- name: qed

# RDK descriptor feature calculator.
- name: rdkdesc

# inchis.csv

InChi,name
"InChI=1S/C8H9NO2/c1-6(10)9-7-2-4-8(11)5-3-7/h2-5,11H,1H3,(H,9,10)","paracetamol"
"InChI=1S/C13H18O2/c1-9(2)8-11-4-6-12(7-5-11)10(3)13(14)15/h4-7,9-10H,8H2,1-3H3,(H,14,15)","ibuprofen"

$ chemfeat calculate features.yaml inchis.csv features.csv

# features.csv

InChi,qed__ALERTS,qed__ALOGP,qed__AROM,qed__HBA,qed__HBD,qed__MW,qed__PSA,qed__ROTB,rdkdesc__FpDensityMorgan1,rdkdesc__FpDensityMorgan2,rdkdesc__FpDensityMorgan3,rdkdesc__MaxAbsPartialCharge,rdkdesc__MaxPartialCharge,rdkdesc__MinAbsPartialCharge,rdkdesc__MinPartialCharge,rdkdesc__NumRadicalElectrons,rdkdesc__NumValenceElectrons
"InChI=1S/C8H9NO2/c1-6(10)9-7-2-4-8(11)5-3-7/h2-5,11H,1H3,(H,9,10)",2,2.0000999999999998,1,2,2,151.16500000000002,52.82000000000001,1,1.2727272727272727,1.8181818181818181,2.272727272727273,0.5079642937129114,0.18214293782620056,0.18214293782620056,-0.5079642937129114,0,58
"InChI=1S/C13H18O2/c1-9(2)8-11-4-6-12(7-5-11)10(3)13(14)15/h4-7,9-10H,8H2,1-3H3,(H,14,15)",0,3.073200000000001,1,2,1,206.28499999999997,37.3,4,1.2,1.7333333333333334,2.1333333333333333,0.4807885019257389,0.3101853515323108,0.3101853515323108,-0.4807885019257389,0,82

Python API¶

Documentation

from chemfeat.database import FeatureDatabase
from chemfeat.features.manager import FeatureManager

# A list of feature specifications, either loaded from a user-provided
# YAML file or set programmatically.
feat_specs = [
    {'name': 'qed'},
    {'name': 'rdkdesc'},
    {'name': 'rdkfp', 'size': 2048}
]

# An iterable of InChi strings, such as a column from a loaded CSV file.
inchis = [
    "InChI=1S/C8H9NO2/c1-6(10)9-7-2-4-8(11)5-3-7/h2-5,11H,1H3,(H,9,10)",
    "InChI=1S/C13H18O2/c1-9(2)8-11-4-6-12(7-5-11)10(3)13(14)15/h4-7,9-10H,8H2,1-3H3,(H,14,15)"
]

# Create the database object. This can point to a common database to centralize
# caching of calculated features.
feat_db = FeatureDatabase('features.sqlite')

# Create the feature manager object.
feat_man = FeatureManager(feat_db, feat_specs)

# Calculate the features and retrieve them as a Pandas dataframe.
feat_dataframe = feat_man.calculate_features(inchis, return_dataframe=True)

ChemFeat Conclusion¶

Already supports a wide range of feature sets.
Easy to add new feature sets.
Simple command-line tool and Python API.
Available on PyPI.

MolPred¶

Combine Hydronaut and ChemFeat to predict molecular properties using ML & DL.

Hydronaut¶

An ML & DL framework for managing and optimizing hyperparameters with Hydra and tracking results with MLflow

Presentations and tutorials.

Recent feature: Python decorators!

MolPred Overview¶

Trains ML & DL models using feature vectors generated by ChemFeat.

Highlights¶

All the benefits of Hydronaut (parameterized via YAML configuration files, automatic hyperparameter sweeping and optimization, systematic result tracking with MLFlow.
All of the benefits of ChemFeat (large number of features, easily extensible to add more feature sets).
Automatic visualization of numeric and categoric features. Examples
Easy to define user models and scorers.

Basic Usage¶

Define a model as a subclass of ModelBase.
- ModelBase subclass example
Modify the configuration file template.
- Configuration file example
Provide a CSV file with InChis, prediction targets and optional additional features.
Run hydronaut-run.

MolPred Conclusion¶

Relatively simple to adapt a model to the framework.
Everything can be configured with YAML (model, feature sets, various metrics).
Easy to explore & optimize hyperparameters thanks to Hydronaut (test different models, feature sets, etc.).
Automatic visualization of features.
All runs are tracked (parameters, metrics, models, artifacts, plots, code version, etc.)
Retrieve models via MLflow for testing and prediction.

The End¶

Back to homepage