Research

Learning latent variables driving cancer

Cancer represents a complex ecosystem with a patient’s individual genomic aberrations, immune infiltrate, and transcriptional programs culminating in a unique disease. Despite this, there are overarching traits present across cancers described first in 2000 with Hanahan and Weinberg’s first The Hallmarks of Cancer paper, followed by The Hallmarks of Cancer: The Next Generation in 2009 and Hallmarks of Cancer: New Dimensions in 2022. By developing novel therapies to block these axes, like avoiding immune destruction and sustaining proliferative signaling, cancer biologists have come to understand that single molecule biomarkers fail to capture the nuanced microenvironmental conditions necessary for therapeutic response.

My work uses matrix factorization to identify key gene expression programs in squamous cell carcinomas and further characterize the corresponding latent biology using neural networks to learn factors from alternative data types (mutation, histology).

Exploring biomarker detection from tumor histology

Hematoxylin and eosin (H&E) stained slides are collected from each cancer patient for their diagnosis. Despite their ubiquitous clinical usage, recent studies have demonstrated that there is additional prognostic information encoded in this data modality that can be revealed using deep learning. Prior work from Pearson lab has demonstrated that genomic features are detected in tumor histology. While genomic alterations themselves can't be detected in an H&E slide at this resolution, the genomic aberrations likely reflect a latent feature of the tumor, as do histologic features, which is why you can predict mutation from histology.

I am using a Generative Adversarial Network to learn histologic features that cause a classifier to flip the predicted class. The features learned from this process shed light on the types of histologic features that are relevant for genomic classification problems.

Prior work

Before starting graduate school I worked as a research technician in a cancer biology lab at the University of Chicago studying the immune modulating effects of targeted therapies for prostate cancer.

As an undergraduate I participated in NSF funded Research Experiences for Undergraduates (REUs) at the Harvard-Smithsonian Center for Astrophysics, National Oceanic and Atmospheric Administration (NOAA), and University of California, Los Angeles (UCLA) applying my education in mathematics and physics to understand solar wind, space weather, and particle laden flows.