Research Project Title:
Machine Learning and Metabolomics for Disease Diagnosis
abstract:Medical misdiagnosis is a pervasive and costly problem, especially when diagnostic tests are invasive or even nonexistent. We aim to improve the efficacy of blood tests, which today rely on searching for individual biomarkers. We propose instead to use the entire molecular signature as input to a machine learning model for diagnosis. If successful, we would be able to provide faster diagnoses at a much lower cost.
A wealth of relevant data exists, but true clinical application will require algorithms which can handle high-dimensional data, control for confounding factors, and remain robust to shifts in data distribution over time. We aim not only to develop such robust algorithms, but also to work with medical collaborators to evaluate such a model's real world performance.
"I'm excited to use my math background to take theoretical results and adapt them for real life applications."