Data Packages for the Dissertation Repository: Generative Models for de novo Molecular Design

The data/ directory contains the input data, intermediate files, and processed outputs used in the dissertation evaluation workflow. Structure:bootstrap_threshold_analysis/: intermediate bootstrap outputs used for effect-size threshold estimation for scaffold-based and pharmacophore-based metrics.comparison_outputs/: outputs from scaffold-based versus pharmacophore-based comparison, including overlap analysis, UMAP projections, cross-modality analysis, and derived figures.enamine.smi: SMILES file used for the Enamine baseline/reference set. fine_tuning_comparison/:outputs used for the comparison of pretrained and fine-tuned generator variants.generator_statistical_testing/: outputs from pairwise statistical testing of generators, including q-value matrices, significance matrices, and derived heatmaps for scaffold-based and pharmacophore-based evaluation.generators_without_finetuning/: generated sets and derived outputs used for comparison between pretrained and fine-tuned generators.information_about_clusters/: metadata and summary statistics describing cluster structure for each receptor.input_recall_sets/: recall sets used as reference inputs for metric calculation.nuclear_receptor/: source data related to the nuclear receptor case study (Glucocorticoid_receptor).output_sets/: generated molecule sets and derived workflow inputs.protease/: source data related to the protease case study (Leukocyte_elastase).results_pharm_based/: final pharmacophore-based metric outputs.results_scaffold_based/: final scaffold-based metric outputs.subset_analysis/: outputs from the Output Set subset-size stability analysis for scaffold-based and pharmacophore-based metrics.thresholds_ph4/: outputs from pharmacophore similarity-threshold sensitivity analysis, including intermediate results and visualizations. All analysis scripts are available in the associated GitHub repository: GitHub link. Note: precomputed RDKit pharmacophore fingerprint files were not included in the Zenodo deposition because of their large size. These files can be regenerated locally using the compute_pharmacophore_fingerprints.py script available in the associated GitHub repository.