Research
Featured Work
Proteomic Biomarkers of Biological Aging and Frailty Trajectories
People of the same chronological age decline at very different rates. This research develops circulating protein signatures trained directly on the pace of functional decline rather than on age itself, and asks whether such measures capture biological aging and forecast downstream health. ProFPS, a 112-protein plasma signature trained on longitudinal change in body weight, gait speed and handgrip strength in the Framingham Heart Study, transfers without refitting to FHS Generation 3 and the UK Biobank, where it tracks functional aging and predicts incident age-related disease and mortality.
👉 ProFPS on GitHub
Frailty and Cognitive Aging: Shared Mechanisms
This research investigates the interconnected biological mechanisms driving both physical frailty and cognitive decline, including chronic inflammation, immune dysfunction, and metabolic dysregulation. Using AI/ML tools and multi-omics integration, we aim to identify mechanisms and pathways that explain the shared underpinnings between frailty phenotypes and neurodegenerative processes.
Inflammation, Immunosenescence, and Cognitive Aging
This research series investigates how peripheral inflammation and immunosenescence contribute to cognitive impairment and dementia, using data from the Framingham Heart Study. Through a combination of inflammatory protein profiling, immune cell phenotyping, and brain imaging, these studies have advanced our understanding of systemic aging and brain health.
👉 Explore Project
Multimodal Learning Across Molecular Networks and Clinical Trajectories
Molecular measurements and clinical phenotypes carry complementary information but differ in structure, one being a network and the other a trajectory over time. This line of work develops deep learning architectures that fuse these modalities rather than modelling either alone, with emphasis on interpretability, identifying which molecules, time points and cross-modal interactions drive a prediction. MGRFusionNet, the current framework, couples a graph neural network protein encoder with a recurrent model of clinical trajectories through a set-transformer fusion layer.
👉 MGRFusionNet on GitHub
ONDSA: A Testing Framework for Omics Network Structures Comparison
A novel statistical testing framework for identifying differential and similar omics network structures across multiple groups based on sparse Gaussian Graphical Models.
👉 Read more
Transfer Learning for High-Dimensional Network Data
The target domain typically carries far fewer samples than the source, and the two can differ in two distinct ways: the regression model itself may shift, and so may the dependence structure of the underlying network. This work develops transfer learning theory and estimators for high-dimensional graph and network convolutional regression that accommodate both model shift and dependence shift, characterizing when borrowing strength from a source domain improves estimation and node classification in the smaller target sample.
👉 Graph convolutional regression (arXiv) · Network convolutional regression (arXiv)
swdpwr: Power Calculation for Stepped Wedge Cluster Randomized Trials
An R package and SAS macro for power/sample size calculation of stepped wedge cluster randomized trials (SWCRT).
👉 View swdpwr Documentation
More projects are under construction and will be available shortly.