Research with the Penn Medicine BioBank

Statistical genetics and machine learning on one of the largest academic biobanks in the US

Role: researcher · University of Pennsylvania · 2024 – present
Links: Penn Medicine BioBank

In one paragraph

The Penn Medicine BioBank (PMBB) links genetic data to years of electronic health records for hundreds of thousands of Penn Medicine patients. I use it extensively for biological research: I design and run analyses that combine statistical genetics and machine learning to find the genes and variants behind hearing loss, cardiovascular disease and breast cancer.

The biobank in numbers

Participants enrolled	260,000+
Participants of non-European ancestry	~30%
Average electronic health record history per participant	~7 years
Data linked to each participant	genotypes, exome sequences, EHR (diagnoses, labs, medications)

_{Public figures from Penn Medicine. The size of each study’s cohort is reported in its own publication.}

Research areas

Hearing loss

Rare-variant gene-burden tests and exome-wide association studies (ExWAS), in collaboration between the Hall and Epstein labs. The work started as a replication of an earlier analysis and grew into a pipeline designed to be rerun in other biobanks, so findings can be validated independently. I present the results to the research groups.

Cardiology

Machine-learning models trained on GPU clusters to study the genetic architecture of cardiovascular traits, alongside association analyses.

Methods

• Statistical genetics: gene-burden tests, ExWAS and GWAS, with multiple-testing correction
• Machine learning on HPC and GPU clusters
• Scale: billions of genetic variants and multi-terabyte datasets
• Reproducibility: pipelines built to be rerun in other biobanks

Stack

Python · HPC/GPU clusters · statistical genetics · ML

_{Methods and research questions only. No participant-level data or unpublished results are shared.}