Tools & Projects

HLA Matchability Calculator

Public Django web application on AWS EC2 — real-time individualized HLA matching-opportunity assessment against 2.17M NMDP nine-locus reference genotypes; PDF / CSV / spreadsheet export. The first U.S. individualized HLA matching-opportunity metric, analogous to CPRA on the matching-likelihood side.

Transplant Toolbox

Modernized the Gragert Lab's suite of public-facing transplant immunology tools: upgraded the Django stack and dependencies, redesigned the frontend, and deployed updates to production (AWS EC2 via Bitnami).

Originally developed by the Gragert Lab, Tulane University School of Medicine.

HaploFreqImpute

High-resolution HLA imputation model leveraging expectation-maximization, ~10M NMDP haplotype frequencies, and linkage-disequilibrium patterns to resolve intermediate-resolution typing ambiguity into calibrated allele-level posterior probabilities at HLA-A, -B, -C, -DRB1, -DRB3/4/5, -DQA1, -DQB1, -DPA1, -DPB1.

AlleleFreqImpute

Single-locus baseline imputation pipeline; comparator to HaploFreqImpute.

Probabilistic Offer Filter

Reference implementation demonstrating automation of the UNet match-run via imputed allele-level HLA typing at the 1% DSA-probability threshold. Simulation demonstrates reduction in manual histocompatibility review and expansion of the compatible donor pool. Brier / TRS calibration via the Paynter et al. open-source framework.

Clinical Typing Parsers

Three production-grade ingestion pipelines that convert heterogeneous vendor-specific histocompatibility-laboratory typing outputs into structured GL Strings consumable by the imputation pipeline: PDF-to-structured-data extractor; RT-PCR SureTyper XML-to-GL-String parser; nine-locus LinkSeq XML-to-GL-String parser. Deployed by the UPenn / Kamoun group for multi-laboratory deceased-donor typing extraction.

HLA-RAG

Domain-specific retrieval-augmented generation system over ~3,500 curated HLA articles; exposed to LLMs via Model Context Protocol for citation-grounded scientific querying through any LLM (ChatGPT, Claude, Gemini).

ILC3-RAG

Retrieval-augmented generation system built on ~2,600 curated Q1 journal articles focused on ILC3 biology, lung infection, and immune response. Literature processed, tokenized, and embedded into a vector database, then connected to LLMs via MCP for rapid, comprehensive, hallucination-minimized literature access.