Overview
The biology of human tissues underlies how cells interact, adapt, and change over time. In cancer and aging, this biology becomes profoundly disrupted: tissues lose their organized architecture, immune regulation falters, and RNA regulatory networks shift. These changes drive disease initiation, progression, and therapeutic resistance. Yet, directly decoding tissue biology has remained challenging, especially in the clinical context, where the richest archives of patient samples have been historically inaccessible to advanced molecular profiling.
Our group develops and applies technologies to bridge this gap. We (1) combine microsystem engineering and biochemical innovation to advance spatial multi-omics technologies and apply them to clinically archived human tissues, transforming decades of pathology samples into spatially resolved molecular atlases; (2) build AI-driven computational pipelines to integrate multi-omics data with histopathology and clinical records, enabling the computation of quantitative indicators and the discovery of spatially enriched regulators linked to patient outcomes; and (3) use in vitro and in vivo systems to model these pathways, allowing deep mechanistic studies, biomarker discovery, therapeutic hypothesis generation, and the development of new treatment strategies.
Following the ‘3D’ pathway—Decode, Design, and Develop—we bring together technology development, clinical translation, and therapeutic design into a unified framework. In the face of unforeseen challenges, our multifaceted approach ensures we can pivot, adapt, and continue pushing the boundaries of spatial omics and cancer and aging biology.
Our Pipelines
Omni-Spatial Toolkit: Decoding High-Performance Multi-Omics from Clinically Archived FFPE Samples
Formalin-fixed paraffin-embedded (FFPE) tissues are essential in clinical practice, being the backbone of histopathology diagnoses. Pathology departments have accrued vast collections of FFPE blocks over time, creating a rich, yet underutilized compendium of materials that, accompanied by clinical data, stands as a treasure trove for human tissue biology research. Bai Lab leads in the innovation of spatial-omics technologies spanning whole-transcriptome, epigenome, and functional genomics, turning decades of archived pathology samples into spatially resolved molecular atlases directly linked to clinical outcomes. These platforms enable comprehensive, multimodal sequencing—at base-pair resolution—of matched tumor and normal tissues, or sections collected before and after immunotherapy, to uncover mechanisms of tumorigenesis and therapeutic response. Applied longitudinally, they further provide clinically relevant insights into the molecular underpinnings of healthy versus pathological aging.
We are developing AI-integrated computational pipelines to decode the regulatory axes underlying spatial heterogeneity and to nominate actionable targets or vulnerabilities within specific tumor regions or cell types. Our approaches include variational autoencoders and deep canonical correlation analysis for multimodal data fusion, graph neural networks to identify driving molecular features, and interpretable machine learning frameworks to rank candidate regulators for target prioritization. By integrating spatial multi-omics with AI, we leverage region-specific dysregulation to uncover mechanistic insights, define clinically relevant biomarkers, and inform therapeutic strategies.
AI-Integrated Framework for Identifying Spatial Regulatory Drivers
In Vitro and In Vivo Models to Drive Translational Therapeutics
Building on targets identified through spatial multi-omics and AI integration, we will employ in vitro and in vivo models to evaluate their therapeutic potential. Patient-derived organoids and co-culture systems will be used to functionally validate candidate regulators and assess mechanisms of action. Promising targets will then be advanced into in vivo mouse models—including orthotopic and humanized systems—to test efficacy, specificity, and safety. This pipeline establishes a direct path from computational discovery to mechanistic validation and preclinical therapeutic development.