Single-cell & Spatial Omics Analysis
Single-cell RNA sequencing (scRNA-seq) enables the identification of rare cell types and cell type-specific gene expression patterns by measuring individual cells rather than population averages. We develop statistical methods for:
- Cell type annotation and clustering: Cellcano for scATAC-seq cell typing, FEAST for feature selection, and WIND/CEMUSA for clustering evaluation.
- Differential expression analysis: SC2P uses mixture models (zero-inflated Poisson and lognormal-Poisson) for flexible differential expression inference in sparse scRNA-seq data.
- Spatial transcriptomics: SIGLE for spot-level representation learning, STAND for abnormality detection, and MicroMap for predicting spatial expression from H&E images.
Signal Deconvolution for Bulk Omics Data
Bulk omics data remain the primary choice for population-level studies due to lower costs. We develop methods to extract cell type-specific information from bulk data—effectively "unmixing the smoothie":
- Signal deconvolution: Estimating cell type proportions and reconstructing pure cell type profiles without single-cell reference data.
- Cell type-specific inference: Accounting for cell type mixture in differential expression/methylation and sample clustering.
Software: TOAST (reference-free deconvolution) and InfiniumPurity (tumor purity estimation).
Differential Analysis for Bulk Omics Data
We develop rigorous statistical methods for detecting differential signals in bulk sequencing data:
- BS-seq: DSS uses hierarchical beta-binomial models for differential methylation analysis (20,000+ annual downloads).
- RNA-seq: Methods accounting for overdispersion and small sample sizes.
- ChIP-seq: Differential peak detection for protein-DNA binding sites.