Longitudinal imaging
Near-lifespan tracking of brain microvasculature
Integrated optical imaging and automated analysis to quantify vascular morphology, topology, and flow over seven months.
Nature Communications ↗Research / Projects
Selected work in multimodal learning, precision oncology, therapeutic discovery, and quantitative biomedical imaging.
Research / Projects
Machine learning for biological translation
My current research focuses on multimodal and self-supervised learning for therapeutic discovery and precision medicine. I work with tumor molecular profiles, functional-genomic screens, and drug-response data to predict cancer dependencies and prioritize therapeutic targets.
Current research
Multimodal AI for therapeutic discovery
Functional-genomic screens reveal which genes cancer cells depend on, while pharmacologic screens measure how they respond to treatment. Because these experiments are performed primarily in cell lines and other model systems, translating their results to patient tumors remains a central challenge. I develop multimodal learning approaches that combine these experimental measurements with tumor molecular profiles to identify vulnerabilities in patient tumors and prioritize therapeutic targets.
Featured public project
MANTIS
Methylation profiling and Artificial Neural networks for Time-resolved Individualized Survival predictions
A multimodal, biologically guided deep-learning framework for individualized survival and treatment-effect prediction in childhood brain tumors.
MANTIS integrates genome-wide DNA methylation, copy-number, and clinical information using sparse connectivity priors and a discrete-time survival model. Rather than assigning patients to broad risk groups, it produces individualized, time-resolved outcome estimates and supports treatment-intensity stratification.
The public repository includes code for model training, inference with pretrained weights, evaluation, and visualization.
Earlier work
Quantitative imaging and physiological prediction
My doctoral and earlier research combined deep learning, signal processing, simulation, and optical imaging to measure biological systems more accurately.
Deep learning + simulation
Red blood cell flux estimation
Trained a one-dimensional convolutional neural network on simulated time series to estimate capillary red blood cell flux from optical coherence tomography.
Frontiers in Neuroscience ↗3D image analysis
OCT microangiogram analysis toolbox
Developed deep-learning methods for automated enhancement, segmentation, and graph-based vascular quantification.
Biomedical Optics Express ↗ · GitHub ↗Quantitative imaging
Confocal profile and focal-plane mapping
Created a method to estimate tilted or curved focal planes for more accurate OCT attenuation-coefficient mapping.
Biomedical Optics Express ↗Clinical signal processing
Consciousness indexing with resting-state EEG
Combined quantitative EEG features and machine learning for consciousness assessment and outcome prediction.
Brain Topography ↗