Â鶹ӳ»­

Modeling of T cell receptors (TCRs), T cell antigens, and their interactions using AI

Identifying which antigens are recognized by which TCRs is one of the most fundamental unsolved problems in immunology and a prerequisite for rational T cell-based therapy design. The lab has made sustained contributions to this problem through a series of AI models of increasing scope and capability. The team published pMTnet (Nature Machine Intelligence, 2021), the first deep learning model to predict TCR¨CpMHC binding specificity from CDR3¦Â sequence, antigen sequence and MHC allele alone, achieving AUROC of 0.833 on held-out data and revealing that human endogenous retrovirus E antigens are more immunogenic than canonical neoantigens in kidney cancer.

The lab subsequently developed pMTnet-omni (Nature Communications, 2026), a hybrid sequence-structure model that extends prediction to both MHC class I and II, and to both human and mouse TCR-pMHC pairs. They showed that pMTnet-omni is capable of therapeutic TCR engineering with a lab-in-the-loop procedure. Complementing these binding models, the lab developed TESSA (Nature Methods, 2021), which integrates TCR sequences with single-cell RNA-seq data to map the functional landscape of TCR repertoires, demonstrating that TCR sequence similarity constrains T cell phenotype and dictates a gradient in antigen targeting efficiency across clonotypes. The team further developed Netie (Nature Methods, 2022), a model to infer the evolutionary dynamics of neoantigen-T cell interactions within tumors over time, and the CSiN score (Science Immunology, 2020), which incorporates neoantigen clonality and immunogenicity to predict immunotherapy outcomes. Together, this body of work has established the Tao Wang Lab as a leading group in AI-driven modeling of T cell antigen recognition.

Modeling of B cell receptors (BCRs)/antibodies, B cell antigens, and their interactions using AI

B cell and antibody biology presents a distinct set of computational challenges from TCR modeling: BCR sequences are shaped by somatic hypermutation, antigen-driven selection, and class switching, requiring models that integrate receptor sequence, clonal evolution, and single-cell phenotype. The lab has developed a series of AI approaches to address these challenges. The team published Benisse (Nature Machine Intelligence, 2022), a contrastive learning model that jointly embeds BCR sequences and single B cell gene expression profiles, enabling the first systematic study of how BCR sequence shapes B cell functional state across diseases. Applying Benisse to COVID-19 patient cohorts, the team demonstrated that BCR signaling-dependent rearrangement is most pronounced in severe disease phases and resolves upon recovery, providing mechanistic insight into humoral immune dynamics during infection. The team also developed BepiTBR (iScience, 2022), a linear B cell epitope prediction model based on T-B cell reciprocity, showing that explicitly incorporating predicted HLA class II epitope enrichment, particularly for DQ allele binders, substantially improves B cell epitope prediction accuracy, with implications for vaccine design. Most recently, the team developed Cmai (Nature Cancer, 2025), a deep learning framework that profiles BCR-antigen binding affinities from tumor-infiltrating B cell repertoires and demonstrated that Cmai-derived BCR affinity scores in tumor-associated B cells predict immune checkpoint inhibitor treatment outcomes. This work establishes that the antigen-binding landscape of the tumor B cell compartment carries clinically actionable information.

Methodological development for single-cell sequencing and spatially resolved transcriptomics data analysis

The rapid adoption of single-cell and spatial profiling technologies has created an urgent need for specialized computational methods that can handle their unique data structures and biological complexity. The lab has made consistent contributions in this space across the full arc of these technologies. The team published SCINA (Genes, 2019), among the first semi-supervised cell typing algorithms for scRNA-seq and CyTOF data, enabling biologically guided cell annotation without unsupervised clustering. The lab developed Sprod (Nature Methods, 2022), a graph-based denoising algorithm for spatially resolved transcriptomics that integrates spatial coordinates and histological image features to recover low-expression signals lost to technical noise, substantially improving downstream analyses. The team subsequently developed Spacia (Nature Methods, 2024), a multi-instance learning framework to infer directional cell-to-cell interactions from single-cell resolution spatial transcriptomics data, which the lab validated in multiple biological contexts, including uncovering ligand-receptor interaction programs that drive cell-2-cell crosstalk in the spatial tissue microenvironment. For mass cytometry, the lab developed Cytomulate (Genome Biology, 2023) and comparative benchmarks of dimensionality reduction methods (Nature Communications, 2023), providing the field with validated computational standards for CyTOF data analysis. Very recently, the lab developed SPACER for decoding the rules of cellular recruitment in solid tissues from spatially resolved transcriptomics data (Nature Biomedical Engineering, 2026). Dr. Wang also published a Science Immunology commentary in 2026 to discuss about the development of spatial immune cell receptor profiling technologies from Dr. William Hudson and other investigators in the field.

Collaborative bioinformatics research

A major dimension of the Tao Wang Lab's work is providing bioinformatics leadership for collaborative translational studies in which computational analyses are central to the scientific discovery. These collaborations leverage the tools and models developed in the lab and extend them to address specific disease questions with clinical relevance. In prostate cancer, Dr. Wang collaborated with the Ping Mu Laboratory to analyze multi-omic data revealing that ZNF397 deficiency triggers TET2-driven epigenetic reprogramming and lineage plasticity, providing a mechanistic basis for androgen receptor-targeted therapy resistance (Cancer Discovery, 2024). In liver biology, Dr. Wang contributed to two high-profile studies with the Hao Zhu laboratory: one identifying that PKD1 mutant hepatic clones suppress steatohepatitis without promoting cancer (Cell Metabolism, 2024), and a landmark study characterizing the metabolic zonation-dependent origin of hepatocellular carcinoma (Science, 2026). In the TCR-epitope prediction space, Dr. Wang co-led the development of GTE, a graph learning framework for TCR-epitope binding specificity prediction, published in Briefings in Bioinformatics (2024). Across oncology more broadly, Dr. Wang's neoantigen and tumor microenvironment analysis pipelines have been adopted by multiple groups at UT MD Anderson and UT Southwestern, contributing to publications on brain tumors, lung cancer, kidney cancer, and melanoma. The Tao Wang Lab also contributes on an ongoing basis to the Database of Actionable Immunology (DBAI), a centralized computational portal that hosts the team's models and pipelines for broad community use.