Research
My research develops statistical methods for high-dimensional and structured data, with an emphasis on scalable inference, interpretability, and uncertainty quantification. I am particularly interested in problems where complex structure arises naturally—such as spatial, network, and multi-modal data—and in designing computationally efficient models that remain scientifically meaningful.
Spatial Transcriptomics
Spatial transcriptomics integrates gene expression with spatial information to study tissue organization and heterogeneity. My work focuses on Bayesian models for joint inference of spatial structure and gene-level variation.
JASPER: Joint Bayesian Analysis of Spatial Expression via Regression
Dey, P., Guhaniyogi, R., Ni, Y., Mallick, B.K.
arXiv preprint. Submitted.Joint Integrative Spatial Transcriptomics via Bayesian Modeling for Domain Recovery and Spatially Variable Gene Selection
Dey, P., Guhaniyogi, R., Ni, Y., Mallick, B.K.
In preparation.
Variational Inference
I develop scalable approximate Bayesian inference methods for complex likelihoods. A central theme is tangent approximation, a computationally efficient alternative to classical variational methods with strong theoretical grounding.
- A Generalized Tangent Approximation Framework for Strongly Super-Gaussian Likelihoods
Roy, S., Dey, P., Pati, D., Mallick, B.K.
arXiv preprint. Major revision, JASA (Theory and Methods). Code.
Symbolic Regression
Symbolic regression aims to discover interpretable mathematical expressions directly from data. My work focuses on Bayesian and variational approaches for structured, uncertainty-aware symbolic learning.
VaSST: Variational Inference for Symbolic Regression using Soft Symbolic Trees
Roy, S., Dey, P., Mallick, B.K.
arXiv preprint. Submitted. Code.Hierarchical Bayesian Operator-induced Symbolic Regression Trees for Structural Learning of Scientific Expressions
Roy, S., Dey, P., Pati, D., Mallick, B.K.
arXiv preprint. Submitted. Code.
Multi-Omic Integration
My work in bioinformatics focuses on integrating multiple omics modalities for biomarker discovery and mechanistic understanding. Key themes include representation learning, modality prioritization, and mediation analysis.
Noninvasive fecal multi-omic signatures for ApcS580/+/KrasG12D/+ mice
Ivanov, I., Mullens, D., Dey, P., Chung, H.C., Gaynanova, I., Ni, Y., Ufondu, A., Yang, F., Han, H., Woods, P., Goldsby, J.S., Davidson, L.A., Safe, S.H., Jayaraman, A., Chapkin, R.S.
In preparation.Diet shapes gut microbial function and indirectly influences host gene expression in infants
Dey, P., Mullens, D., Ivanov, I., Chapkin, R., Donovan, S., et al.
In preparation.
Neuroimaging
Neuroimaging data involves high dimensionality, multi-resolution structure, and complex interactions across modalities. My work develops scalable models for joint analysis of network and spatial imaging data.
Additive Nonparametric Regression with Spatial and Network Objects
Guhaniyogi, R., Dey, P., Chandra, K., Scheffler, A., Mallick, B.K.
Major revision, Biometrics. Code.Outlier detection for multi-network data
Dey, P., Zhang, Z., Dunson, D.B.
Bioinformatics, 2022. DOI. Code.Ensembles of Mondrian Processes for Continuous Modeling of Structural Connectomes
Dey, P., Zhang, Z., Dunson, D.B.
In preparation.Hierarchical Ensembles of Mondrian Processes for Simultaneous Modeling of Common and Individual Structure of Connectome Data
Dey, P., Zhang, Z., Dunson, D.B.
In preparation.
Machine Learning for Causal Inference
I have worked on scalable matching-based methods for causal inference, with emphasis on interpretability and computational efficiency in large datasets.
- dame-flame: a Python package providing fast interpretable matching for causal inference
Gupta, N.R., Orlandi, V., Chang, C.-R., Wang, T., Morucci, M., Dey, P., Howell, T.J., Sun, X., Ghosal, A., Roy, S., Rudin, C., Volfovsky, A.
Journal of Statistical Software, 2025. DOI. Code.