Curated AlphaFold-Derived Structural Feature Dataset for Rare Disease Variant Classification

Overview This repository contains a curated, feature-engineered dataset developed to support research into the use of AlphaFold protein structure predictions for early-stage rare disease variant interpretation. The dataset was constructed by integrating multiple open-access biological resources and transforming them into a unified, machine-learning-ready format. The dataset is intended to support reproducible research into structure-aware pathogenicity prediction and AI-assisted diagnostic decision support within HealthTech contexts. Data Sources All raw data used in constructing this dataset were obtained from publicly available, open-access biological databases: ClinVarSource of clinically interpreted genetic variants and pathogenicity labels.Variants classified as Pathogenic, Likely pathogenic, Benign, or Likely benign were used. AlphaFold Protein Structure Database (AlphaFold DB)Source of predicted three-dimensional protein structures and per-residue confidence scores (pLDDT). Original data ownership remains with the respective providers. This dataset represents a derived and transformed research artefact, not a redistribution of raw source data. Feature Engineering and Dataset Construction The final dataset consists of engineered numerical features designed to capture structural confidence and predicted perturbation effects associated with genetic variants. Key feature engineering steps included: Protein-level structural confidence extraction Average pLDDT score pLDDT standard deviation Counts of high-confidence and low-confidence residues Protein-level confidence ratios Variant-level structural impact estimation Heuristic estimation of variant-specific pLDDT values Penalisation of terminal residues to reflect known AlphaFold uncertainty Differential perturbation modelling for pathogenic versus benign variants Derived comparative features Difference between variant-level and protein-average confidence Normalised confidence deviation metrics All features were generated programmatically using Python and standard scientific computing libraries. The dataset was cleaned, filtered, and validated to ensure consistency and suitability for supervised machine learning. Licensing and Data Ethics This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. No patient-identifiable or sensitive personal data are included. All source databases are publicly accessible and ethically approved for research use. Users are required to cite the original data providers when reusing or extending this dataset. The dataset complies with research data governance principles, including transparency, reproducibility, and responsible reuse. Intended Use This dataset is provided strictly for research and educational purposes. It is intended to support: Methodological research in AI-assisted genomics Evaluation of AlphaFold-derived structural features in variant classification Development of explainable machine learning models for rare disease research The dataset must not be used for direct clinical diagnosis or decision-making without appropriate clinical validation and regulatory approval.