About

Hanlee Ji is an Associate Professor of Medicine (Oncology) and, by courtesy, of Electrical Engineering at Stanford University. He is affiliated with the Center for Artificial Intelligence in Medicine & Imaging (AIMI). His research focuses on the application of artificial intelligence and imaging technologies in medicine, particularly in oncology. As a faculty member at Stanford, he contributes to advancing AI-driven healthcare solutions and collaborates across disciplines to improve medical diagnostics and treatment. His role involves both research and education, supporting initiatives that integrate AI with clinical practice.

Research topics

• Genetics
• Biology
• Computational biology
• Cancer research
• Immunology
• Evolutionary biology
• Cell biology
• Ecology
• Botany

Selected publications

• DNA conjugation on functionalized plastic surfaces for sequential, iterative single molecule sequencing

  Scientific Reports · 2026-01-28

  articleOpen access

  We developed a method for repeated and sequential retrieval of arbitrary DNA elements. A click chemistry process was used to conjugate the DNA molecules onto a plastic surface within the interior of microcentrifuge tubes. For this study, we utilized synthetic DNA sequences that encode arbitrary data and designed PCR primers for amplification. Specifically, the DNA was tailed with trans-cyclooctene (TCO) and was then conjugated to plastic surfaces functionalized with methyltetrazine (MTz). The covalent DNA attachment to the plastic surface enables repeated and non-destructive polymerase-based copying and amplification of the original source molecules. In summary, we demonstrate a new type of DNA storage media with the property of long-term stability and ability to read different groups or files of DNA data. For this proof-of-concept study, we demonstrate the key features of this technology including: (1) characterization of the DNA conjugation process using control strands, (2) conjugation kinetics, and (3) amplification and sequencing of specific DNA data elements over multiple retrieval operations from the same plastic surface. Specifically, we showed that different DNA "file groups" with information could be accessed multiple times with single molecule sequencing. We also compared the performance metrics of PCR versus recombinase polymerase amplification (RPA). Our results indicate that RPA has better performance metrics in terms of reducing contamination and improved yield of retrieved data from across sequential experiments. Overall, we demonstrated a plastic-based DNA storage system for robust and reliable iterative, targeted DNA retrieval, and sequencing.

  PublisherOA PDFDOI

• Abstract 5918: Single cell functional characterization of cancer mutations and their cellular phenotype

  Cancer Research · 2026-04-03

  articleSenior author

  Abstract Many cancer driver mutations lead to proteins with amino acid substitutions which dramatically affect their function. This class of cancer mutations play a critical biological role in cancer progression and maintenance of nearly all malignancies. However, the effect of substitution mutations is mostly inferred computationally rather than functionally tested. As a result, there is practically no biological information about their functional consequences for these substitutions except for a small number.We developed a high-throughput single cell approach to systematically investigate the functional effects of reported cancers substitutions. This system provides parallel, highly scalable testing of many cancer mutations across multiple genes in a single experiment. It uses CRISPR base editors to introduce specific cancer mutations into the genome, identifies the newly introduced mutation genotype among individual cells and determines each mutation’s transcriptional phenotype per a given cell. Specifically, long-read targeted sequencing is applied to single cell cDNAs. Single cell long read sequencing identifies the presence of an engineered mutation in each cDNA assigned to an individual cell. To determine phenotype of the mutation, we integrate the short-read transcriptome profile from the same single cells. This integrative approach enables single-cell direct genotyping of the introduced genetic variant and matching phenotype from the same cell.We chose a set of mutations that occur with high frequency as reported in the TCGA’s pan-cancer atlas. All the possible gRNAs were designed based on the location of the C/A bases in the oncogenic mutations’ sequence context. We detected the engineered mutations by linking actual mutation genotype determined by long-read sequencing with corresponding transcriptome change detected by short-read sequencing in one single experiment.These results demonstrate how combining single cell genomics and direct genome engineering method increase the scale for characterizing diverse cancer-associated mutations. In the future, we will evaluate in parallel large sets of different cancer mutations, how they alter gene expression and changes in the cellular states. Importantly, characterizing the function of novel oncogene mutations may lead to the discovery of new targeted therapeutics for cancers. Citation Format: Huiyun Sun, Dongin Lee, HoJoon Lee, Susan M. Grimes, Raegan Wood, Hanlee P. Ji. Single cell functional characterization of cancer mutations and their cellular phenotype [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 5918.

  PublisherDOI

• Abstract 5491: AI foundation model for single cell annotation from conventional histopathology images of cancer.

  Cancer Research · 2026-04-03

  articleSenior author

  Abstract Spatial imaging technologies have revolutionized our understanding of the tumor microenvironment (TME) by delineating its molecular and cellular architecture. Some of these approaches characterize tissue at single cell resolution. These platforms enable precise cell type annotation and, when spatially registered, allow molecular label transfer onto conventional histopathology images with hematoxylin and eosin (H&E) staining. However, spatial imaging approaches require complex instrumentation and have a high-cost per assay. These limitations prevent the application of spatial analyses on large sets of cancers. In contrast, conventional cancer histopathology slides are widely available and can be imaged at low cost. However, single cell identification from these images remains a manual, semi-quantitative process that is difficult to scale up.To address these limitations, we developed an AI foundation model that enables single-cell characterization directly from conventional H&E images. Our approach uses either spatial proteomic data (e.g., immunohistochemistry - IHC) or spatial transcriptomic profiling. We use these spatial assays to generate molecularly defined labels for diverse cell populations, including epithelial, lymphocyte, macrophage, and stromal cells. For this study, the molecular labels were the basis for training a spatial multimodal classifier. There are two tiers, (1) single-cell H&E crops were embedded using the foundation model H-optimous to obtain high-dimensional representations, and (2) a Multi-Layer Perceptron (MLP) neural network was trained on those embeddings for supervised cell type classification.We trained our model on a set of colorectal cancers (CRC), consisting of 40 multiplexed IHC slides and five Xenium spatial transcriptomic slides. Overall, we had 34 million single cells for model training. The model achieved an overall accuracy of 87.1% and a macro-average area under the receiver operating characteristic curve (AUROC) of 96.1%. Independent validation on seven CRC samples (∼8 million cells) yielded consistent performance with 87.2% accuracy and 95% macro-average AUROC.In summary, our multimodal model enables automated and scalable single-cell-level cell type annotation directly from H&E images. This approach provides a quantitative foundation for immune-tumor interactions for colorectal cancer. Furthermore, by integrating H&E-derived cell type maps with tumor-specific genomic alterations from matched TCGA datasets, the framework enables systematic analysis of important TME cell types such as tumor-infiltrating lymphocytes and their spatial cellular distributions, offering insights into colorectal cancer microenvironmental architecture. Citation Format: Xiangqi Bai, HoJoon Lee, Xiao Tan, Chaoyi Li, Anuja Sathe, Yan Wang, Quan Nguyen, Hanlee P. Ji. AI foundation model for single cell annotation from conventional histopathology images of cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 5491.

  PublisherDOI

• Abstract 4130: Single-cell multi-omic characterization of gastric intestinal metaplasia reveals potential genetic, epigenetic and isoform signatures of lesions at high risk for GC progression

  Cancer Research · 2026-04-03

  articleSenior author

  Abstract Worldwide, gastric cancer (GC) is the fifth most common malignancy and the third leading cause of cancer-related deaths. Gastric intestinal metaplasia (GIM) is a precursor lesion of GC. However, only a small number of GIM lesions progress to GC. A key challenge is identifying the genomic, molecular and cellular features of GIM that predict their risk of becoming invasive cancer. Once these features are identified, one could “intercept” patients at high risk for developing GC. We conducted a single-cell multi-omic analysis of endoscopic biopsies of GIM lesions. These patients underwent GC risk evaluation using a two-biopsy approach. We used the pathology results to determine a clinical risk stage based on the Operative Link on Gastric Intestinal Metaplasia Assessment (OLGIM). Clinical stages include I and II which are low risk versus III and IV which are high risk. Each biopsy underwent single cell RNA-seq and single cell assay for transposase accessibility of chromatin (ATAC). Therefore, we had both of these genomic readouts for each cell. In addition, we used single cell long read sequencing to identify transcript isoforms and mutations. Overall, we obtained the following single cell cell features of GIM which included: (1) gene expression, (2) chromatin accessibility, (3) copy number aberrations, (4) somatic variants, and (5) isoform expression. We compared the single cell genomics features between high-risk and low-risk groups. For example, we observed that key genes specifically expressed in intestinal-like stem cells, which are linked to potential gastric cancer-initiating cells—such as CDH17, SI and CPS1—were more highly expressed among the high-risk group, which also exhibited increased chromatin accessibility at the same genes. Furthermore, we detected differential isoform usage of gastric-related genes between the two groups, which implies that isoform changes are related to GC progression. These findings provide genomic, molecular and cellular features of GIMs that are associated with high risk of developing GC. Citation Format: Dongin Lee, Xiangqi Bai, Sue Grimes, Kyungtae Lee, Yan Wang, Charlotte Wong, Anuja Sathe, Ignacio Wichmann, Youlim Kim, Rithika Meka, Renee Long, Allison Im, Billy Lau, Robert Huang, Hanlee P. Ji, . Single-cell multi-omic characterization of gastric intestinal metaplasia reveals potential genetic, epigenetic and isoform signatures of lesions at high risk for GC progression [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4130.

  PublisherDOI

• Abstract 4133: Single cell multi-omics enables molecular dissection of gastric cancer subtypes

  Cancer Research · 2026-04-03

  articleSenior author

  Abstract Gastric cancer (GC) has different histopathological subtypes that include intestinal, diffuse, and mixed. These subtypes are identified based on histologic cell morphology and are clinically informative, more so than molecular subtypes. Namely, histologic subtypes differ in prognosis, rates of metastasis and treatment response. However, many questions remain about the genomic features that distinguish each histologic subtype. Integrating single-cell datasets is particularly challenging at the molecular level due to technical variation. We developed an integrative data set combining in-house and public multi-omics datasets - they include bulk sequencing, scRNA-seq, spatial transcriptomics, and proteomics. The scRNA-seq data included 208 tumors from 118 patients. This single cell data was harmonized to reduce the effects of batch variability. For analysis, we developed a computational pipeline incorporating generalized linear models, non-negative matrix factorization for meta-program identification, sample-level pseudo-aggregation and network-based gene expression analysis. This pipeline enabled gene-wise dissection, pathway-centric interpretation of malignant transcriptional programs and systematic quantification of cell state dynamics across integrated cohorts. We identified distinct tumor-intrinsic gene expression programs across subtypes. The intestinal and mixed tumors expressed immune-associated programs, while the MSI tumors exhibited elevated cell-cycle signaling compared to MSS. In the tumor microenvironment, the diffuse subtype showed reduced B-cell differentiation but increased dendritic cell abundance, whereas intestinal subtype was enriched for T-cell subtypes, including regulatory T cells with immunosuppressive potential. From the single cell results, we determined that TIGIT expression was significantly elevated among CD8 and regulatory T cells. This result was confirmed with multiplexed immunofluorescence staining on an independent set of 142 GCs. Fibroblast-driven interactions dominated diffuse tumors, while intestinal and mixed tumors displayed immune-mediated “hot tumor” phenotypes. Overall, our study showed that the analysis of an integrated, harmonized single-cell data from GC revealed malignant and microenvironmental programs distinguishing subtypes. Our results provided some subtype-specific vulnerabilities and provides insight into potential future therapeutic targets. Citation Format: Junha Cha, Anuja Sathe, Yan Wang, Susan M. Grimes, Hanlee P. Ji. Single cell multi-omics enables molecular dissection of gastric cancer subtypes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4133.

  PublisherDOI

• Basal gland localization and focal distribution of OLFM4-expressing cells in increasing severity of gastric intestinal metaplasia

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-05-20

  articleOpen access

  ABSTRACT Patients with gastric intestinal metaplasia ( GIM ), a precancerous lesion, are at high risk for progressing to gastric cancer. Identifying these patients is critical to enable gastric cancer interception. Current approaches rely primarily on histologic evaluation of GIM severity and extent, which may be improved by incorporating molecular features that distinguish high-risk lesions. Our prior single-cell and spatial transcriptomics study identified differentially expressed genes associated with the highest-risk category of GIM. They included ANPEP expressed in enterocytes and CPS1 and OLFM4 expressed in intestinal stem-like or progenitor cells. We evaluated the protein expression and localization of these three markers to understand the cellular features associated with GIM risk and their spatial distribution within metaplastic tissues. Using multiplex immunofluorescence, whole slide image analysis and confocal microscopy, we examined protein expression from 100 tissue biopsies annotated for metaplasia severity using the Operative Link on Gastric Intestinal Metaplasia Assessment ( OLGIM ) system. Tissue samples included control gastric tissue, GIM, dysplasia and adenocarcinoma. Quantitative whole slide image analysis demonstrated that CPS1 expression had a modest association with disease severity. Although ANPEP was strongly associated with GIM severity, it was also frequently expressed in stromal regions outside epithelial glands. In contrast, OLFM4 expression was largely restricted to epithelial glands and showed a strong association with increased OLGIM severity. These OLFM4-positive epithelial cells were present in discrete glandular foci that expanded with increasing severity of metaplasia. Within individual metaplastic glands, OLFM4 expression was highest at the gland base with decreased expression toward the gland surface. Overall, these findings identified OLFM4 as a protein marker associated with high-risk GIM. The spatial organization of OLFM4-expressing cells at the base of metaplastic glands and their focal expansion within tissues suggest the presence of a stem cell-like epithelial compartment that may contribute to the progression of GIM towards gastric cancer.

  PublisherOA PDFDOI

• The Gastric Cancer Genome Explorer: A spatial and single cellular registry for identifying novel cellular markers and targets.

  Journal of Clinical Oncology · 2026-01-10

  article1st authorCorresponding

  429 Background: The Gastric Cancer Registry (GCR) is a comprehensive resource that collects clinical information and samples from gastric cancer (GC) patients. The cancer samples undergo extensive molecular and cellular analysis that includes genomic sequencing, single cell and spatial analysis. Afterwards, this data is released through an online resource called the GCR Genome Explorer (GCR-GE) (https://gcregistry-explorer.stanford.edu/). We have added a single cell and spatial analysis data. Methods: The registry’s GC samples have been analyzed with whole genome, whole exome, and bulk RNA sequencing. As part of the latest data release, we have added several hundred GCs that have undergone either single cell or spatial transcriptomics analysis. We collected single cell RNA-seq (scRNA-seq) data from 208 samples from 118 patients. These data sets were generated in-house and downloaded from public datasets. Next, we developed a computational pipeline to mitigate batch effects in scRNA-seq data. This step involved applying generalized linear models, considering network properties and using sample level pseudo-aggregation. Finally, we used non-negative matrix factorization (NMF) to discover gene expression signatures characterizing the GC histologic subtypes. Results: We identified a diverse range of cell types, single cell gene expression profiles and cellular architecture in the context of the native tumor tissues. We focused on the cancer epithelial cells. The intestinal histologic subtype had highest overexpression of the Claudin gene family and JUN - FOS signaling. The diffuse histologic subtype had genes MUC1 and TFF1 were overly represented in diffuse subtype. Expression profiles were validated in an independent set of different GCs that consisted of 64 diffuse, 180 intestinal and 19 mixed cancers. The single cell data provides insight into the tumor microenvironment (TME). The diffuse subtype had higher prevalence of antibody secreting B cells, while intestinal subtype was enriched for pro inflammatory tumor associated macrophages. The intestinal subtype had increased regulatory T cells as an immunosuppressive mechanism. Diffuse GC had similar degree of CD8+ T cell abundance as intestinal GC. Using multiplex immunofluorescence, we identified TIGIT expression in CD8 and regulatory T cells in an independent cohort of 142 GC patients. Conclusions: By defining molecular and cellular characteristics across gastric cancers, this resource will accelerate research for identifying new treatment options such as distinct targetable features across different GC subtypes.

  PublisherDOI

• Abstract 4089: Immunosuppressive cellular topography and genomic adaptations sustain colorectal cancer metastasis to the brain

  Cancer Research · 2026-04-03

  articleSenior author

  Abstract Colorectal cancer metastases to the brain (CRC-BMets) are lethal and resistant to immunotherapy and radiation. The cellular and molecular adaptations that sustain CRC-BMets remain poorly defined. The objective of our study was to identify the tumor-intrinsic and microenvironmental programs that support metastatic growth in the brain. To understand these mechanisms as they operate in the patient’s metastatic niche, we performed sequencing and imaging-based spatial transcriptomics on 51 patients, including a subset with paired primary tumors and longitudinal radioresistant tumors. Using patient-derived three-dimensional co-culture systems, microfluidic assays and single-cell sequencing, we experimentally perturbed the intercellular interactions that promote tumor growth in the metastatic niche.Metastatic tumor cells showed significantly elevated chromosomal instability and activation of RNA-processing, stress-response, and junctional remodeling pathways compared with paired primary tumors. After radiation therapy, resistant clones retained their copy-number alterations and displayed increased epithelial-mesenchymal transition and transcriptional plasticity. Spatial mapping revealed that tumor cells proliferated preferentially in endothelial-rich regions. Tumor cells acted as multicellular ligand hubs (MIF, GDF15, PRSS3, SEMA3C) reinforcing tumor-supportive interactions with multiple surrounding microenvironmental cell types in the metastatic niche. Adjacent brain parenchyma expressed astrocytic and glial activation, together with angiogenic and matrix-associated pro-tumor features. CRC-BMets reconstructed a stromal microenvironment dominated by macrophages and fibroblasts in proximity to tumor cells, with sparse lymphocytes. This conserved macrophage-fibroblast neighborhood promoted angiogenesis and extracellular matrix remodeling, with increased regulatory T cell and exhaustion gene signatures. Macrophages expressed high levels of SPP1, predicted to orchestrate a matrix remodeling immunosuppressive program via interactions with corresponding receptors on neighboring cells. In a microfluidic device, CRISPR deletion of SPP1 reduced macrophage mobility in the presence of tumor cells. When co-cultured with a CRC-BMet patient-derived spheroid and fibroblasts, SPP1 knockout led to reduced expression of lipid-metabolism related genes in macrophages and disrupted tumor-promoting interactions. Together, these findings indicate that CRC-BMets are maintained by spatially organized tumor-intrinsic adaptations and multicellular stromal programs that persist after radiotherapy. These tumor genomic adaptations, multicellular ligand hubs, and an SPP1-dependent macrophage-fibroblast axis define targetable vulnerabilities in this aggressive metastatic site. Citation Format: Anuja Sathe, Mengrui Zhang, Xiangqi Bai, Ji In Kang, Rithika Meka, Huiyun Sun, Mouhita Humayun, Xun Wang, Susan M. Grimes, Aparajita Khan, Mingen Liu, Andrew S. Luksik, Michael Lim, Claudia K. Petrisch, Christopher M. Jackson, Hannes Vogel, Jeanne Shen, Roger D. Kamm, Melanie Gephart, Summer Han, Hanlee P. Ji. Immunosuppressive cellular topography and genomic adaptations sustain colorectal cancer metastasis to the brain [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4089.

  PublisherDOI

• Abstract 1407: Single-cell long-read transcriptomics reveals isoform centric molecular features of rare pseudomyxoma peritonei disease.

  Cancer Research · 2026-04-03

  articleSenior author

  Abstract Pseudomyxoma peritonei is a rare mucin-producing cancer, most often arising from appendiceal mucinous neoplasm. In the past, investigations into transcriptomic and genomic alterations in this tumor have concentrated on gene-level changes, neglecting the impact of differential isoform usages resulting from alternative splicing and allele specific expression. In this study, we conducted single-cell long-read sequencing on 10 biopsies from 5 tumor patients at various stages of disease. Single cell analysis delineated the individual cell types and long read sequencing revealed the sequence of mRNA transcripts from its 5’ transcript start sites to 3’ polyA- tail - providing insights into disease and cell type specific RNA transcript structures and changes. Across the single cell results, we identified a total of 35,095 expressed genes and 130,272 transcripts, including 50,330 novel transcripts. Based on the identified gene and isoform repertoire, epithelial cells derived from appendiceal mucinous neoplasm showed an upregulation of genes and isoforms linked to goblet cell differentiation, cell proliferation, and mucin production. Furthermore, we identified alternative RNA alternative splicing in genes related to protein degradation and cellular transport within tumor epithelial cells. Our ongoing research focuses on discovering phased somatic mutations at the allele-specific isoform level and uncovering RNA splicing-derived neoantigens expressed in tumor epithelial cells. These efforts may reveal novel oncogenic drivers and potential therapeutic targets for this tumor. Citation Format: Kyungtae Lee, Anuja Sathe, Raegan Wood, Dongin Lee, Billy Lau, Hanlee P. Ji. Single-cell long-read transcriptomics reveals isoform centric molecular features of rare pseudomyxoma peritonei disease [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1407.

  PublisherDOI

• Cancer subclone detection based on DNA copy number in single-cell and spatial omic sequencing data

  Nature Methods · 2025-09-01 · 6 citations

  article
  PublisherDOI

Recent grants

• NEUROPATHOLOGY/TISSUE BANK CORE

  NIH · $16.0M · 2021–2026

• NIH Grant U01CA176299

  NIH · $3.6M · 2017

• Organoid-based Discovery of Oncogenic Drivers and Treatment Resistance Mechanisms

  NIH · $4.7M · 2017–2023

• Single Cell Transcriptomic and Genetic Diversity by Single Molecule Long Read Sequencing

  NIH · $259k · 2011–2017

• Single Cell Transcriptomic and Genetic Diversity by Single Molecule Long Read Sequencing

  NIH · $3.7M · 2011–2025

Frequent coauthors

• Susan M. Grimes

  Stanford University

  150 shared

• Billy T. Lau

  Stanford University

  143 shared

• HoJoon Lee

  127 shared

• Anuja Sathe

  Stanford University

  94 shared

• James M. Ford

  University of California, San Francisco

  85 shared

• John Bell

  Stanford University

  73 shared

• Stephanie Greer

  Stanford University

  61 shared

• Joshua D. Schiffman

  Huntsman Cancer Institute

  59 shared