Biological systems are complex, dynamic, and spatially organized. Our research develops computational methods and AI models to understand how biological systems are organized, evolve, and respond to perturbations across molecular, cellular, spatial, and temporal scales.
We seek to move beyond static measurements toward computational models that can represent biological organization, reconstruct tissue structure, and ultimately predict how biological systems change over time.
We develop computational methods for extracting biological structure from single-cell and spatially resolved molecular measurements. Our work addresses cell identity, cellular heterogeneity, spatial organization, and molecular variation across tissues and biological conditions.
Biological measurements are high-dimensional observations of underlying cellular and molecular systems. We develop representation learning approaches that capture relationships among genes, cells, tissues, and biological contexts, with the goal of building more generalizable models of biology.
Most molecular assays provide snapshots of biological systems, while biological processes unfold continuously through space and time. We aim to develop computational models that reconstruct spatial organization and infer how cells and tissues transition between biological states.
We are exploring computational approaches that connect biomedical data with clinical development. A central goal is to build models and simulators that can use real-world evidence and prior clinical knowledge to understand trial populations, treatment effects, and sources of trial failure.
Our long-term goal is to build computational models of biological systems that can reconstruct, simulate, and predict biological states across space and time.
We envision models that integrate molecular measurements, cellular organization, tissue architecture, and temporal dynamics to create computational representations of biological systems. Such models could provide a foundation for simulating biological processes and exploring how tissues respond to perturbations.
Our current work builds on a long-standing foundation in statistical genomics. Earlier research established statistical frameworks for single-cell omics data analysis, differential analysis, signal deconvolution, and modeling heterogeneous molecular data, providing methodological foundations for modern computational biology.
Our most widely used method series, built on empirical Bayes shrinkage of dispersion parameters. The RNA-seq dispersion estimator was adopted by DESeq2, one of the most widely used differential expression tools, and the framework was extended to BS-seq for differential methylation analysis at both single-locus and region level.