How irfanhaider71/graph-stack works
Graph-Stack is a research workflow for predicting physicochemical properties of oncology-relevant compounds from molecular representations. Compound SMILES and property measurements form the input data; degree-based indices are either computed analytically or estimated from molecular descriptors, then used for property prediction. Similarity-based analog groups support grouped validation, while direct descriptor, fingerprint, and neural models provide comparisons. The README also documents Y-scrambling and applicability-domain analysis. The graph combines README-level capabilities with sampled source; unsampled subsystem wiring is shown only where explicitly documented.
Main entry points: dataset smiles.py Compound SMILES · results Prediction and metrics · compute_analytical_indices.py Analytical indices · stacking_property_prediction_analytical_indices.py Analytical-index prediction
Components 19
Compound data
- Compound SMILES
data/dataset smiles.py - Property data
data/build_physicochemical_properties_csv.py - Analog groups
data/AnalogGroup.py
Index generation
- Analytical indices
code/compute_analytical_indices.py - Estimated indices
code/compute_extimated_indices.py - Index surrogate models
code/compute_extimated_indices.py
Prediction and validation
- Analytical-index prediction
code/stacking_property_prediction_analytical_indices.py - Estimated-index prediction
code/stacking_property_prediction_estimated_indices.py - Grouped validation
code/stacking_property_prediction_analytical_indices.py
Comparisons and analysis
- Classical baselines
code/classical_baselines_analytical_indices.py - RDKit and Morgan baselines
code/direct_rdkit_morgan_baselines.py - GIN baseline
code/Graph Isomorphism Network (GIN) baseline.py - SMILES-CNN baseline
code/SMILES-CNN baseline.py - Descriptor MLP baseline
code/Descriptor-based MLP baseline.py - Similarity analysis
code/similarity_grouping_dendrogram_threshold_analysis.py - Index correlations
code/correlation coefficients using analytical .py - Robustness analyses
code - Prediction and metrics
results
Other components
- Researcher
Connections 31
- Researcher → Compound SMILES: provides
- Researcher → Property data: provides
- Compound SMILES → Analytical indices: describes molecules
- Compound SMILES → Index surrogate models: descriptor input
- Index surrogate models → Estimated indices: predicts indices
- Compound SMILES → Analog groups: fingerprint input
- Analytical indices → Analytical-index prediction: index features
- Estimated indices → Estimated-index prediction: index features
- Property data → Analytical-index prediction: target values
- Property data → Estimated-index prediction: target values
- Analog groups → Grouped validation: group labels
- Grouped validation → Analytical-index prediction: validates
- Grouped validation → Estimated-index prediction: validates
- Analytical indices → Classical baselines: index features
- Estimated indices → Classical baselines: index features
- Compound SMILES → RDKit and Morgan baselines: molecular input
- Compound SMILES → GIN baseline: graph input
- Compound SMILES → SMILES-CNN baseline: sequence input
- Compound SMILES → Descriptor MLP baseline: descriptor input
- Compound SMILES → Similarity analysis: similarity input
- Analytical indices → Index correlations: index values
- Analytical-index prediction → Prediction and metrics: writes predictions
- Estimated-index prediction → Prediction and metrics: writes predictions
- Classical baselines → Prediction and metrics: writes metrics
- RDKit and Morgan baselines → Prediction and metrics: writes metrics
- GIN baseline → Prediction and metrics: writes metrics
- SMILES-CNN baseline → Prediction and metrics: writes metrics
- Descriptor MLP baseline → Prediction and metrics: writes metrics
- Similarity analysis → Analog groups: informs grouping
- Researcher → Robustness analyses: requests analysis
- Robustness analyses → Prediction and metrics: records outputs