How can machine learning reveal disease-relevant protein function?
Developing protein-language-model approaches to identify microbial enzymes, antigens, and other biologically meaningful proteins at scale.
Internal Medicine Resident · Physician-Scientist
I am an Internal Medicine resident and physician-scientist at Stanford Health Care, training through the ABIM Research Pathway with plans to specialize in Rheumatology. My research combines machine learning, human immunology, and clinical data to understand immune-mediated disease and develop more precise approaches to diagnosis and treatment.
I earned my MD from the University of Colorado School of Medicine and my PhD from the University of Cambridge, where I was a Gates Cambridge Scholar. Before medical school, I was a software engineer at Uber, an experience that shaped my interest in finding meaningful signals within complex systems and helped draw me from computation toward medicine and biological discovery.
My Cambridge PhD in Biotechnology and Statistics/Deep Learning used protein language models to discover microbial enzymes and microbial protease allergens implicated in immune-mediated disease. I was advised by Dr. Sergio Bacallado at Cambridge and Dr. Ramnik Xavier at the Broad Institute and Mass General Hospital.
My newest paper, Identifying microbial protease allergens through protein language model-guided homology, was published in Cell Systems (2026). It introduces a deep learning framework using protein language models to uncover candidate allergenic serine proteases across gut and oral microbiome gene catalogs. The work was recently profiled by the Cambridge Department of Chemical Engineering and Biotechnology. Other recent work includes CAZyLingua, a protein language model-based tool for annotating carbohydrate-active enzymes in metagenomics (BMC Bioinformatics, 2025).
I am interested in using machine learning, high-dimensional human data, and mechanistic immunology to understand pathogenic immune responses in autoimmune disease and identify clinically actionable biomarkers and therapeutic targets.
I am grateful to the Gates Cambridge Scholarship and Rotary International Scholarship for funding my PhD.
Developing protein-language-model approaches to identify microbial enzymes, antigens, and other biologically meaningful proteins at scale.
Studying the cellular and molecular programs that shape immune recognition, inflammation, and tissue injury.
Connecting mechanistic models with patient phenotypes, biomarkers, therapeutic response, and target discovery.
At the University of Colorado School of Medicine, I earned Honors in all six core clinical clerkships — Internal Medicine, OB/GYN, Pediatrics, Family Medicine, Psychiatry, and Surgery — as well as in my Rheumatology and Medicine Acting Internship rotations. I was elected to Alpha Omega Alpha (AOA), the national medical honor society.
I am now an Internal Medicine resident at Stanford Health Care, training through the ABIM Research Pathway with planned subspecialty training in Rheumatology at Stanford. My long-term goal is to combine rigorous clinical training with research in autoimmune disease as a physician-scientist.
ABIM Research Pathway · Planned Rheumatology training at Stanford
Gates Cambridge Scholar
Uber
San Francisco, CA
University of Colorado Boulder
A deep learning framework using protein language models to identify candidate allergenic serine proteases across gut and oral microbiome gene catalogs.
Thurimella, K., Wu, E., Li, C., Graham, D. B., Owens, R. M., Plichta, D. R., Sokol, C. L., Xavier, R. J., & Bacallado, S. (2026). Identifying microbial protease allergens through protein language model-guided homology. Cell Systems, 0, 101510.
CAZyLingua is the first annotation tool to use protein language models for accurate classification of carbohydrate-active enzyme families and subfamilies in metagenomics.
Thurimella, K., Mohamed, A. M., Li, C., Vatanen, T., Graham, D. B., Owens, R. M., La Rosa, S. L., Plichta, D. R., Bacallado, S., & Xavier, R. J. (2025). Protein language models uncover carbohydrate-active enzyme function in metagenomics. BMC Bioinformatics, 26(285).
A pilot randomized controlled trial showing that AI-enabled tailored messaging increases engagement in cardiovascular health interventions.
Xia, A.†, Thurimella, K.†, Bull, S., Waughtal, J., Chavez, C., Novins-Montague, S., Silvasstar, J., Salyers, A., Ho, M. P., & Lavieri, M. (2024). Pilot randomized controlled trial: Increased engagement in cardiovascular health through AI-enabled tailored messaging. JMIR Cardio (Under Review). †Co-first authors
SCNIC is open-source software that can generate correlation networks and detect and summarize modules of highly correlated features.
Shaffer, M.†, Thurimella, K.†, Sterrett, J. D., & Lozupone, C. A. (2023). SCNIC: Sparse Correlation Network Investigation for Compositional Data. Molecular Ecology Resources, 23(1), 312–325. †Co-first authors