Gene Variant Pathogenicity Prediction and Repair

Live-Demo Link: PRISMATIC

Web-deployed ML platform for computational genetic variant pathogenicity screening and repair

Overview and Usage

PRISMATIC is a full-stack ML application allowing users to input genetic mutation data in the following form:

Chromosome Number: integers 1-24 where X = 23 and Y = 24

Reference Sequence (Original DNA Allele) e.g. "ATCG"

Alternate Sequence (Variant DNA Allele) e.g. "ACAT"

Flank 1 ("Left" side of the alleles)

Flank 2 ("Right" side of the alleles)

PRISMATIC will take the input strings and output:

• Pathogenicity probability and classification status (Pathogenic / Benign)
• Potential therapeutic candidates generated using my algorithm "ReGen"

Model Performance

• 90.0% ROC-AUC
• 89.7% PR-AUC
• 82.0% F1
• 80.6% Pathogenic F1

Stack

• Backend: Python / FastAPI, deployed on Hugging Face Spaces (Docker)
  • ML: Scikit-learn, XGBoost
  • Algorithm logic and feature engineering pipeline source code in GEM / PRISM (see GitHub profile)
• Frontend: React, CSS, deployed on Vercel

Background

• Derived from GEM and PRISM (see GitHub). Built to explore bioinformatics + ML integration and its potential in the future of medicine.

Disclaimer: Not intended for clinical use

This project is intended for educational and research purposes only. This is not a medical tool and should not be used for clinical decision-making.