This platform represents a clinical-grade precision medicine system designed to provide deep biological insights into polygenic disease susceptibility. By integrating high-throughput genomic data with functional biological databases, the system moves beyond simple risk scoring to offer a comprehensive interpretation of genetic variants. The platform utilizes advanced ensemble machine learning, SHAP-based explainability, and force-directed graph theory to trace the causal chain from individual single-nucleotide polymorphisms (SNPs) to systemic pathway disruptions.
The architecture emphasizes a zero-database dependency model, where all genomic reference data (13,918 GWAS records) and biological ontology data (3,464 genes) are loaded into memory at runtime to ensure maximum performance and data privacy.
The system replaces traditional database overhead with a high-speed, in-memory data engine. At startup, the platform ingests a curated dataset of over 13,000 GWAS records, including odds ratios, risk allele frequencies, and genomic coordinates. This allows for near-zero latency cross-referencing of patient genotypes.
The core predictive engine operates through a sophisticated 10-step pipeline:
- Variant Harmonization: Cross-referencing patient rsIDs against the internal GWAS corpus.
- Phenotypic Feature Engineering: Constructing a high-dimensional feature vector based on variant pathogenicity and genomic context.
- Ensemble Model Inference: Utilizing champion models selected through 5-fold stratified cross-validation. Current models include XGBoost, Random Forest, and Logistic Regression ensembles tailored for Type 2 Diabetes, Coronary Artery Disease, Breast Cancer, and Hypertension.
- XAI (Explainable AI): Generating SHAP (SHapley Additive exPlanations) values to decompose the model's decision-making process, providing per-feature impact metrics.
- Pathogenic Variant Ranking: Identifying and ranking individual SNPs by their contribution to the overall risk probability.
- Polygenic Risk Scoring (PRS): Calculating traditional clinical risk metrics as a baseline for comparison with machine learning outputs.
- Pathway Enrichment Analysis: Mapping identified genes to biological pathways to detect systemic functional disruptions.
- Metabolic Kinetic Interpretation: Analyzing enzyme kinetics for genes encoding metabolic proteins.
- Automated Clinical Synthesis: Generating a structured, natural-language narrative that summarizes all findings into a professional report.
- Topological Network Construction: Creating a node-link model for visualization of the variant-gene-pathway hierarchy.
The platform uniquely integrates several authoritative data sources to interpret genetic findings:
- Variant-Disease Associations: Derived from the GWAS Catalog, providing the statistical foundation for risk calculations.
- Pathway Ontologies: Utilizing Reactome data to understand how individual gene mutations converge on shared biological processes.
- Enzyme Kinetics: Integrating BRENDA database parameters (Km values) to evaluate the potential impact of mutations on enzymatic substrate affinity and metabolic flux.
This project utilizes data from several open-access biological databases. Users of this platform must adhere to the respective licensing terms of these organizations.
Pathway data and gene-to-pathway mappings are derived from the Reactome Knowledgebase.
- License: Reactome data is provided under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.
- Citation: Fabregat A, et al. (2018). The Reactome Pathway Knowledgebase. Nucleic Acids Res. 46(D1):D649-D655.
Enzyme kinetic parameters (Km values) are sourced from the BRENDA database.
- License: BRENDA is freely available for academic and non-profit use. Commercial entities may require a license through Bio-Base or the database owners.
- Usage: Data is processed locally and encapsulated within the
biology_reference.jsonfile.
SNP-disease association data is sourced from the NHGRI-EBI Catalog of published genome-wide association studies.
- Usage: Data has been cleaned and filtered for the four target diseases (Type 2 Diabetes, CAD, Breast Cancer, Hypertension).
Developed with FastAPI and Python 3.10+, the backend serves as the primary intelligence orchestrator. It handles model serialization (.pkl), XAI computation (SHAP), and biological enrichment logic. All components are modularized for horizontal scalability of disease models.
The user interface follows a "Scientific Elegance" design philosophy, prioritizing data density and visual clarity.
- Interactive Karyogram: Visualizing SNP distribution across the human chromosomes.
- Risk Architecture: Utilizing animated SVG gauges for risk probability visualization.
- Genomic Vulnerability Network: A force-directed 2D network graph (interactive canvas) showing the biological hierarchy of risk.
- Multi-Disease Radar: A comparative profile showing cross-disease vulnerability based on shared genetic pathways.
- Python 3.10 or higher
- Node.js 18 or higher
- NPM 9 or higher
- Initialize a Python virtual environment:
python -m venv env - Activate the environment:
.\env\Scripts\activate(Windows) orsource env/bin/activate(Linux/Mac) - Install dependencies:
pip install -r backend/requirements.txt - Execute the server:
python backend/main.py
- Navigate to the frontend directory:
cd frontend - Install packages:
npm install - Start the development server:
npm start
Executes the full 10-step intelligence pipeline for a specific patient genotype and target disease.
Generates the graph-theoretical representation of the variant-pathway-disease relationship for the network visualization.
Performs a simultaneous analysis across all supported disease models to identify shared genetic risks.
This platform is intended for research and educational purposes only. The risk predictions generated by the AI models are based on statistical associations in population-level GWAS data and do not constitute a clinical diagnosis or medical advice. Genetic risk assessments should be interpreted by qualified healthcare professionals or genetic counselors. The authors and project participants assume no liability for medical or health-related decisions made based on this software.