My research focuses on applying machine learning (ML) techniques to scientific domains. More specifically, I develop accurate, informative, and interpretable machine learning methods in fields such as pathology, medicine, and biochemistry. I have a passion for using my knowledge of computing to fuel advances in applied sciences and observing the direct impact my research has on a community.
Postdoctoral Research
My current research projects involve using machine learning to learn from medical images as well as relevant pathology and clinical data.
Kidneys
Digital pathology, utilizing whole slide imaging and artificial intelligence, has gained popularity as a method to improve diagnosis and prognosis of kidney disease [5]. I am investigating which factors are important in determining whether to discard or keep donor kidneys. A significant number of kidneys are being discarded that appear unremarkable by histopathologic examination. Using machine learning, we can elucidate other unknown factors leading to organ discard.
Ovaries
Ovarian cancer is one of the primary causes of death among women. Awareness, early detection, and advanced treatments are imperative to combat ovarian cancer and mitigate the devastating impact on women’s lives. I am developing a deep learning model to predict malignancy and categorize ovarian neoplasms from intraoperative frozen section digital whole slide images into three distinct classes: benign, borderline, and malignant.
PhD Research
Throughout my PhD, my research involved using machine learning in a variety of scientific domains [1].
Medicine
I constructed deep learning models to disentangle effects of race and socio-economic and clarify the actual drivers of health disparities are rooted in structural racism [4]. In a related study, I developed an adversarial variational auto-encoder that mitigates the impact of race on latent spaces using clinical data (submitted to Communications Medicine).
I created a machine-learning model to guide patient-specific anesthetic dosages of fentanyl and propofol to prevent post-induction hypotension. This work was presented at the Society for Technology in Anesthesia (STA) 2023 (submitted to Anesthesiology).
biochemistry
I demonstrated that accurate quantum chemical computations could be performed without optimized geometries by operating in the coordinate-free domain using machine learning on graph encodings [3]. In another study, I developed machine learning models to learn higher-level chemical properties from lower-level quantum calculations. [2].
I also examined the trends of existing structural alerts, or known toxic molecular substructures, and whether they are the most effective way for chemists to determine toxicity within drugs. This lead to the development of a subgraph mining algorithm that identifies important substructures associated with bioactivated molecules (submitted to Chemical Research in Toxicology).
Undergraduate Work
For my undergraduate senior research project at Stetson University, I worked on a project that involved optimizing a machine learning algorithm called Classification by Discriminative Interpolation (CDI). CDI is a supervised learning algorithm that performs classification for functional or time series data. After introducing parallelism and exploiting the fact that each class is examined independently, we saw a decrease in execution time of up to 50%.
References
[5] Goodman, K., Sarullo K., Gaut J. P., Jain S., Role of Artificial Intelligence in Kidney Pathology: Promises and Pitfalls, Kidney360, 2024.
[4] Sarullo, K., D. M. Barch, C. D. Smyser, C. Rogers, B. B. Warner, J. P. Philip, S. K. England, J. Luby, and S. J. Swamidass. Disentangling Socioeconomic Status and Race in Infant Outcomes: A Neural Network Analysis, Biological Psychiatry: Global Open Science, 2023.
[3] Matlock, M. K., M. Hoffman, N. L. Dang, D. Folmsbee, L. A. Langkamp, G. Hutchison, N. Kumar, K. Sarullo, and S. J. Swamidass. Deep learning coordinate-free quantum chemistry, Journal of Physical Chemistry A, 2021.
[2] Sarullo, K., M. K. Matlock, and S. J. Swamidass. Predicting small-molecule bioactivity from deep-learning representations of quantum chemistry, Journal of Physical Chemistry Special Issue: Machine Learning in Physical Chemistry, 2020.
[1] Sarullo, K. Translating Deep Learning into Scientific Domains: Case Studies in Quantum Chemistry, Bioactivation, and Clinical Studies, Ph.D. Dissertation, Washington University in St. Louis, 2023.