Statistics for structured data
High-dimensional datasets often have structure we can use: variables may form a network, observations may share spatial context, or a small set of latent factors may explain much of the variation. I develop methods that incorporate this information into estimation and dimension reduction.
My work includes sparse and graph-constrained canonical correlation analysis, topic models, matrix and tensor factorization, and denoising over networks. The aim is to obtain interpretable estimates with statistical guarantees.
Recent work: CCA as reduced rank regression · Tensor topic modeling · Sparse topic modeling
Understanding learning on graphs
Graph neural networks learn from relationships between observations. I study their statistical properties: how graph convolutions transform a signal, how network structure affects prediction, and how to select and evaluate a model when observations are dependent.
This research brings together theoretical analysis, graph signal processing, and methods for uncertainty quantification. Current directions include model selection for unsupervised graph representations, semi-supervised learning, and graph transformers.
Recent work: GCN convolutions in regression · Current preprints
Supported in part by my NSF CAREER project, Towards Responsible Graph Neural Networks (2023–2028).
Data integration & uncertainty
Biological datasets often combine measurements from genomics, transcriptomics, metabolomics, imaging, and environmental conditions. I develop statistical methods to connect these views and identify relationships that would be difficult to see in any one dataset.
Uncertainty quantification is central to this work. Alongside sparse CCA and multivariate regression, recent projects develop conditional conformal prediction methods to assess the uncertainty of model predictions.
Recent work: SpeedCP · Efficient sparse CCA
Biological & scientific applications
Spatial transcriptomics
Modeling spatial gene expression, tissue organization, and cell-state transitions using structured statistical methods.
Plant & microbial systems
Integrating molecular and environmental data to study thermotolerance, microbial interactions, and host–microbiome responses to stress.
Microbial communities
Using network models and latent structure to understand community organization and connect genetic variation to microbial traits.
Networks in public health
Studying partially observed epidemics and heterogeneous transmission.
Earlier work
My earlier research includes brain connectomics and the analysis of functional MRI, network dynamics, cryo-electron microscopy, and statistical modeling for COVID-19. Across these projects, a common question is how to extract reliable information from complex, noisy data.
Research in practice
Inside the SIGNAL Lab
Find our current projects, team, and open-source software on the lab website.