Computational Molecular Medicine

Single-cell multi-omics data present unique challenges: extreme sparsity and technical noise from dropouts, high dimensionality with thousands of features per cell, and pervasive batch and donor effects that can obscure true biological variation. Robust normalization, dimensionality reduction, and noise-modeling approaches are therefore essential to recover reliable signals and ensure that downstream analyses reflect genuine cellular heterogeneity rather than artifacts of the technology. Our lab combines advanced computational methods with high-dimensional biological data to uncover mechanisms of disease, with a particular emphasis on cancer and metabolic disorders. By integrating single-cell multi-omics, and in vivo model data, we aim to build a mechanistic and spatially resolved understanding of disease progression and therapeutic response.

Single-Cell Multi-Omics in Clinical Cohorts

We specialize in the analysis of clinical single-cell multi-omics datasets, particularly from cancer and metabolic disease cohorts. Using state-of-the-art machine learning techniques, we analyze single-cell RNA sequencing and chromatin accessibility data to characterize cellular heterogeneity and regulatory dynamics. Our pipeline includes robust dimensionality reduction, clustering, and batch correction methods, allowing us to identify distinct cell populations and states across individuals. Through probabilistic graphical modeling and motif enrichment analysis, we reconstruct gene regulatory networks that govern disease-specific transcriptional programs. These approaches allow us to overcome the sparsity and noise inherent to single-cell data and extract biologically meaningful patterns that inform prognosis and therapeutic strategy.

Multi-Omics in Mouse Models of Cancer

To explore tumor development and treatment effects in vivo, we analyze multi-omics data generated from mouse models, including xenografts and genetically induced cancers. These datasets are complex, encompassing multiple axes of variation such as treatment regimens, time points, and tumor subtypes. To disentangle these factors, we develop tailored probabilistic latent variable models (LVMs) that reveal how sources of variability interact and which molecular features are relevant to human disease. This framework enables us to translate findings from experimental models into clinically actionable insights.