AI for health and beyond

Generative AI & digital twins

A digital twin is a computational model that mirrors a real person or system closely enough to test what-if scenarios on it. I use generative AI to build these twins and to augment scarce data, for example improving real-time anomaly detection in data from wearable devices. I also study when GenAI-enhanced twins can be trusted, and apply generative models to problems in the physical sciences.

• GenAI-based digital twins aided data augmentation for anomaly detection in wearable data. Sensors, 2025.
• Opportunities and credibility challenges of GenAI-enhanced digital twins. Computing in Science & Engineering, 2026.
• Software: TwinMe4AD

AI for healthcare

Hospital-acquired infections such as MRSA are costly and often preventable. Working with clinicians, I combine electronic health records with patient contact networks to predict who is at risk, trace where infections come from, and support precaution decisions. This theme also covers privacy-preserving health research with federated learning and visual analytics for patient trajectories.

• Prediction of hospital associated infections during continuous hospital stays. AAAI, 2026.
• Improving MRSA risk prediction with network features. JMIR AI, 2024.
• Spectral characterization to identify HAI patients. Scientific Reports, 2023.
• Privacy-first health research with federated learning. npj Digital Medicine, 2021.

ML for molecules & drug discovery

Designing a new molecule usually means many rounds of expensive lab work. I develop models that predict structural and chemical properties directly from a compound's structure, so labs can prioritize the most promising candidates. Current directions include small-molecule discovery and drug repurposing using protein interaction data.

• P2MAT: ML-driven software for property prediction of materials. Journal of Cheminformatics, 2026.
• Polyketide synthase-based synthesis of polycyclopropanated fuel molecules. Nature Communications, 2026.
• Predicting critical micelle concentration of surfactant mixtures. Journal of Cheminformatics, 2025.
• Software: P2MAT

Topological data analysis

My doctoral work developed scalable topological methods, based on the mapper construction, for exploring high-dimensional data. They find subpopulations and paths through the data that standard clustering misses, and I've applied them to crop phenomics, COVID-19 data and clinical cohorts.

• Hyppo-X: a scalable exploratory framework for complex phenomics data. IEEE/ACM TCBB, 2019.
• Interesting paths in the mapper complex. Journal of Computational Geometry, 2019.
• Detecting divergent subpopulations using interesting flares. ACM-BCB, 2018.
• Software: Hyppo-X

Beyond health: physics, cybersecurity & quantum computing

The methods I develop for health data, especially generative models and learning from limited data, carry over to other sciences. I've applied generative machine learning to lattice QCD in particle physics, and I'm exploring machine learning for cybersecurity and quantum machine learning.

• Generative ML for lattice QCD matrix elements. Physical Review D, 2025.