How pharminfovienna/off-target-p-ml works
OfftargetP-ML is a research toolkit for developing binary small-molecule off-target classifiers and predicting activity profiles from molecular structures. The documented workflow starts with compound/target/activity data, curates SMILES and calculates ECFP4 fingerprints, then trains models by target and evaluates their predictions. The repository also includes saved neural-network models and a script for applying models to new structures. The README documents neural-network and random-forest workflows, but their training scripts are absent from this tree; the sampled AutoGluon scripts show training and evaluation code directly. Connections for undocumented or unsampled workflows are therefore omitted or marked unverified.
Main entry points: autogluon_models.py AutoGluon models · NeuralNetworks Neural-network training · Off-targetP_ML.R Structure prediction · Datasets Compound datasets
Components 14
Data preparation
- Compound datasets
Datasets - Fingerprint preparation
- AutoGluon preparation
AutoGluon/autogluon_filesprep.R
Model development
- AutoGluon models
AutoGluon/autogluon_models.py - Auto-Sklearn models
Auto-Sklearn/Auto-Sklearn.py - Neural-network training
NeuralNetworks - Random-forest training
RandomForest
Prediction
- Saved neural models
Models - Structure prediction
Models/Off-targetP_ML.R - Applicability domain
Applicability_domain.R
Assessment
- Neural evaluation
NeuralNetworks/evaluation.R - AutoGluon metrics
AutoGluon/autogluon_models.py - Auto-Sklearn metrics
Auto-Sklearn/Auto-Sklearn.py
Other components
- Researcher
Connections 15
- Researcher → Compound datasets: provides data
- Compound datasets → Fingerprint preparation: curates and encodes
- Fingerprint preparation → Neural-network training: supplies features
- Fingerprint preparation → AutoGluon models: supplies features
- Researcher → AutoGluon models: runs training
- Researcher → Auto-Sklearn models: runs training
- Researcher → Neural-network training: runs training
- Researcher → Random-forest training: runs training
- AutoGluon models → AutoGluon metrics: calculates metrics
AutoGluon/autogluon_models.py - Auto-Sklearn models → Auto-Sklearn metrics: calculates metrics
Auto-Sklearn/Auto-Sklearn.py - Researcher → Structure prediction: submits structures
- Saved neural models → Structure prediction: loads models
- Structure prediction → Applicability domain: checks applicability
- Researcher → Neural evaluation: evaluates models
- Neural-network training → Saved neural models: produces models