How irfanhaider71/mag-net works
This repository is a reproducibility pipeline for molecular-property prediction: it derives graph descriptors, trains and evaluates the gated multi-task MAG-Net, and compares it with classical and alternative-representation baselines. The documented workflow uses a canonical held-out split and reports cross-validation and robustness results, alongside learned gating/error analysis. Edges reflect explicit README workflow or sampled calls; several README-listed robustness, figure, and correlation scripts are absent from the supplied tree, so their implementation links are not inferred.
Main entry points: baseline_harness.py Shared baseline data · proposed_validated_pipeline.py Validated MAG-Net · topological_descriptors_computation.py Graph descriptor computation · extract_gating_and_errors.py Gating and errors
Components 19
Descriptor analysis
- Graph descriptor computation
1-Topological Descriptor Computation/topological_descriptors_computation.py - Descriptor definitions
1-Topological Descriptor Computation/topological_formulas.py
MAG-Net workflow
- Validated MAG-Net
2-MAG-Net/proposed_validated_pipeline.py - Joint tuning
2-MAG-Net/magnet_joint_tuning.py - Fold summary
2-MAG-Net/magnet_tuned_kfold_summary.py - Gating and errors
2-MAG-Net/extract_gating_and_errors.py
Baseline comparisons
- Shared baseline data
3-Baselines/baseline_harness.py - Classical models
3-Baselines/classical_baselines.py - SMILES CNN
3-Baselines/baseline_smiles_cnn.py - CNN search
3-Baselines/optimize_smiles_cnn.py - Transformer search
3-Baselines/optimize_transformer_smiles.py - Fingerprint search
3-Baselines/optimize_morgan_fp.py - Ungated MLP tuning
3-Baselines/fair_tune_descriptor_mlp.py
Evaluation outputs
- Benchmark metrics
3-Baselines/baseline_harness.py - Robustness studies
2-MAG-Net - Paper figures
2-MAG-Net
Other components
- Researcher
- Molecular data
- Descriptor table
Connections 32
- Researcher → Graph descriptor computation: runs
- Researcher → Validated MAG-Net: runs
- Researcher → Joint tuning: runs
- Researcher → Classical models: runs
- Molecular data → Graph descriptor computation: provides SMILES
- Graph descriptor computation → Descriptor table: computes
- Researcher → Descriptor definitions: views definitions
- Molecular data → Validated MAG-Net: provides inputs
- Descriptor table → Validated MAG-Net: provides descriptors
- Validated MAG-Net → Benchmark metrics: evaluates model
- Molecular data → Joint tuning: provides inputs
- Descriptor table → Joint tuning: provides descriptors
- Molecular data → Fold summary: provides inputs
- Descriptor table → Fold summary: provides descriptors
- Molecular data → Gating and errors: provides inputs
- Descriptor table → Gating and errors: provides descriptors
- Gating and errors → Benchmark metrics: writes analysis
- Molecular data → Shared baseline data: loads and merges
- Shared baseline data → Classical models: supplies data
- Shared baseline data → SMILES CNN: supplies data
- Shared baseline data → CNN search: supplies data
- Shared baseline data → Transformer search: supplies data
- Shared baseline data → Fingerprint search: supplies data
- Shared baseline data → Ungated MLP tuning: supplies data
- SMILES CNN → CNN search: provides tokenization
- Classical models → Benchmark metrics: reports results
- CNN search → Benchmark metrics: reports results
- Transformer search → Benchmark metrics: reports results
- Fingerprint search → Benchmark metrics: reports results
- Ungated MLP tuning → Benchmark metrics: reports results
- Researcher → Robustness studies: runs documented studies
- Researcher → Paper figures: regenerates