hmacdope-omsf/cyp3a4_chembl_lgbm_dev

CYP3A4 ChEMBL pChEMBL regressor

WARNING THIS MODEL IS FOR DEMONSTRATION PURPOSES ONLY

This project demos a machine learning model for predicting CYP3A4 inhibition values (pChEMBL) based on molecular structures. CYP3A4 is a critical enzyme involved in drug metabolism, and understanding a compound's interaction with this enzyme is essential for drug development and safety assessment.

To get started, clone this repo and use the demonstration inference.py script which looks like the below.

Key components:

Architecture: LightGBM gradient boosting regressor Input features:

• ECFP4 fingerprints (Extended Connectivity Fingerprints with radius 4)
• Mordred molecular descriptors

Output: pChEMBL value prediction (higher values indicate stronger inhibition)

from openadmet_models.models.gradient_boosting.lgbm import LGBMRegressorModel
from openadmet_models.features.combine import FeatureConcatenator
from openadmet_models.features.molfeat_properties import DescriptorFeaturizer
from openadmet_models.features.molfeat_fingerprint import FingerprintFeaturizer
import json


model = LGBMRegressorModel.deserialize("model.json", "model.pkl")
featurizer = FeatureConcatenator(featurizers=[FingerprintFeaturizer(fp_type="ecfp:4"), DescriptorFeaturizer(descr_type="mordred")])

# create some sample smiles 
smiles = ["CCO", "CCN", "CCOc1ccccc1"]

# featurize the smiles
X, _ = featurizer.featurize(smiles)

# predict the properties
y_pred = model.predict(X)

print(y_pred)