Artificial intelligence is increasingly ubiquitous in research and medicine. However, proper training and validation of these large-data models are critical for safe adoption in the clinical setting.
I have used multiscale physics-based simulations based on interpretable biological and clinical findings to create high-fidelity data for training neural operators. Such simulations can be varied based on population distributions of vascular geometries, material properties, and hemodynamics to create rich, clinically-relevant datasets at a volume impossible to acquire in vivo. By bridging these biomechanical models with machine learning tools, we can greatly accelerate adoption of artificial intelligence for clinical prediction while maintaining robust standards of data transparency, interpretability, and quality.
My foundation in machine learning was established as an undergraduate at Johns Hopkins University, where I double majored in computer science (with a concentration in computational medicine) and biomedical engineering. At that time, I completed courses in machine learning, computer vision, and surgical robotics, alongside computer science fundamentals. These early experiences gave me a lasting familiarity with computational tools that have continued to shape my research vision as I moved into biomechanics.
As artificial intelligence has become a central topic in many research fields, this background has been extremely valuable for integrating data-driven approaches into my work where they complement mechanistic insight. As I have developed projects at this intersection, it has become clear to me that these approaches can be used jointly to increase the predictive power, precision, and interpretability of both. With these insights in mind, I plan to use machine learning not as a catch-all solution, but as a targeted tool for advancing our understanding of complex biological systems.
Currently, my multiscale work is focused on diving even more deeply into mechanistic models at the cellular level. I have established a gene expression-to-tissue models utilizing RNA-sequencing data for modeling developing pulmonary arteries, with the aim of providing insights into critical age ranges for surgical intervention in patients with congenital heart defects. I have also worked closely with top researchers in the field of artificial intelligence to generate in silico datasets for machine learning that will be translated into the clinical setting to predict disease outcomes. Through these collaborations, I have gained appreciation for how to use multiscale modeling and machine learning to responsibly and impactfully build models capable of advancing cardiovascular medicine.
Postnatal Pulmonary Artery Development From Transcript to Tissue
Schwarz, E. L., Ramachandra, A. B., Yeung, N., Manning, E. P., Weiss, D., & Humphrey, J. D. (2025)
bioRxiv, Under Review at Journal of The Royal Society Interface
Link
Neural Operators for Forecasting Thoracic Aortic Aneurysm Growth and Rupture
Cao, Q. (co-first), Schwarz, E. L. (co-first), Li, D. S., Goswami, S., Means, C. W., Assi, R., Humphrey, J. D., & Karniadakis, G. T. (2025)
In Preperation
Link
Mechanisms Driving Thoracic Aortic Aneurysm Stability
Schwarz, E. L., Li, D. S., C. W. Means, R., Assi, & Humphrey, J. D. (2025)
bioRxiv, Submitted to Biomechanics and Modeling in Mechanobiology
Link
A Cross-scale Causal Mapping Framework Pinpoints Macrophage Orchestrators of Balanced Arterial Development
Han, J., Kong, D., Schwarz, E. L., Takaesu, F., Lee, S., Kim, C., Yang, E., Park, M., Kim, E., Kim, W., Ramachandra, A. B., Manning, E. P., Humphrey, J. D., Park, H., & Davis, M. (2025)
bioRxiv, Under Review at Circulation Research
Link