Statistical Bioinformatics Seminar: Dr Xiaomeng Wan (HKUST)

Title: Zero-Shot Foundation Model for a Universal Gene Expression Atlas of Human Tissue:
Unveiling Clinically Relevant Cell States and Disease-Specific Spatial Niches 

Abstract: The rapid accumulation of single-cell datasets from diverse organs and tissues
presents significant opportunities for understanding complex diseases, yet challenges
remain in effectively analyzing this wealth of information and further leveraging it to
various data types, including spatial transcriptomics (ST) and bulk RNA-seq datasets.
Here, we introduce UniGeneX, a generative single-cell foundation model designed to
reconstruct a universal gene expression profile from extensive transcriptomic data.
UniGeneX minimizes batch effects while preserving biological variability, enabling the
identification of shared gene programs across tumor samples.  By providing consistent
cell type labels and leveraging biological patterns from training data, UniGeneX
facilitates the discovery of disease-specific cell niches in spatial and key cell states
associated with clinical outcomes.  Our model addresses existing limitations in current
single-cell foundation models by focusing on a universal gene expression framework
rather than merely learning embeddings for downstream tasks.  We demonstrate the
effectiveness of UniGeneX in characterizing disease-relevant cell states in glioma and
idiopathic pulmonary fibrosis (IPF), ultimately advancing our understanding of the
mechanisms underlying complex diseases.  

About the speaker: Dr Xiaomeng Wan is currently a Postdoctoral Associate in the
Department of Mathematics at the Hong Kong University of Science and Technology (HKUST),
under the guidance of Prof Can Yang.  She earned her PhD from HKUST under the mentorship
of Prof Can Yang.  Her research centres on statistical machine learning and deep
learning, particularly exploring their applications in the analysis of transcriptomics
datasets.