Dr. Urbanowicz’s research focuses on the development, evaluation and application of bioinformatics and machine-learning methods in biomedical and clinical problems. The motivation for these new methods stems from the challenges presented by (1) large-scale data analysis; (2) integrating data types/sources; (3) complex patterns of association (e.g., epistasis, genetic heterogeneity and pleiotropy; (4) rare variants; (5) genome-to-phenome analysis; (6) the need for interpretable (non-black-box) models; (7) noisy, incomplete data; (8) covariates; and (9) matching or adapting method(s) to a problem with limited domain knowledge. Therefore, his work targets strategies that are scalable, adaptable, interpretable and clinically translatable.

He has led the development of various open-source algorithmic software packages. The GAMETES package was designed to facilitate the simulation of complex genetic models and associated datasets. This makes it possible to rapidly generate a broad-spectrum simulation study for algorithmic testing, evaluation and comparison. The ReBATE package was designed to offer a suite of fast, flexible feature selection methods able to account for complex feature associations. The ExSTraCS package introduced an advanced rule-based machine learning algorithm, otherwise known as a Learning Classifier System (LCS). ExSTraCS was designed to detect, model and characterize disease associations in noisy, complex and heterogeneous data. ExSTraCS was the first and is still the only machine learning method that has directly solved a longstanding computer science benchmark problem characterized by multivariate patterns of feature interactions and heterogeneous patterns of association. Dr. Urbanowicz is one of a handful of international experts on LCS algorithms and co-authored the book Introduction to Learning Classifier Systems in 2017.

He leads the UrbsLab (Unbounded Research in Biomedical Systems), a highly collaborative research group to enable discoveries that help optimize clinical research and patient care through the effective use of high-dimensional, messy and/or complex genomic and clinical data. Collaborative applications have included, but are not limited to, the investigation of obstructive sleep apnea, pancreatic cancer, bladder cancer, congenital heart disease, HIV, falls risk, ER-to-ICU risk prediction, risk of complications following surgical procedures, and transplant rejection.