James Zou

Stanford University
Country flagUnited States

Research Interests

ApplyKite TwinLearns as you apply

Make your next move.

Prepare your application

Review drafts before you send them.

Explore related searches

Articles (18)

Year:

2026

Year:

2026

Recent advances in pathology foundation models (PFMs), which are pretrained on large-scale histopathological images, have significantly accelerated progress in disease-centered applications. In parallel, spatial multi-omic technologies collect gene and protein expression levels at high spatial resolution, offering rich understanding of tissue context. However, current models fall short in effectively integrating these complementary data modalities. To fill in this gap, we introduce spEMO, a novel computational system that unifies embeddings from pathology foundation models and large language models (LLMs) to analyze spatial multi-omic data. By incorporating multimodal representations and information, spEMO outperforms models trained on single-modality data across a broad range of downstream tasks, including spatial domain identification, spot-type classification, whole-slide disease-state prediction and interpretation, inference of multicellular interactions, and automated medical report generation. The outstanding performances of spEMO in these tasks demonstrate its strength in both biological and clinical applications. Additionally, we propose a new evaluation task, known as multi-modal alignment, to assess the information retrieval capabilities of pathology foundation models. This task provides a principled benchmark for evaluating and improving model architectures. Collectively, spEMO represents a step forward in building holistic, interpretable, and generalizable AI systems for spatial biology and pathology.

Year:

2025

The rise of antibiotic-resistant pathogens such as Staphylococcus aureus has created an urgent need for new antibiotics. Generative artificial intelligence (AI) has shown promise in drug discovery, but existing models often fail to propose compounds that are both effective and synthetically tractable. To address these challenges, we introduce SyntheMol-RL, a reinforcement learning-based generative model that can rapidly design synthetically accessible small molecule drug candidates from a massive chemical space of 46 billion compounds. SyntheMol-RL improves upon our prior Monte Carlo Tree Search (MCTS)-based SyntheMol model by generalizing across chemically similar building blocks and enabling multi-parameter optimization. We applied SyntheMol-RL to generate candidate antibiotics against S. aureus by optimizing for both antibacterial activity and aqueous solubility, and we found that SyntheMol-RL generated molecules with improved predicted properties compared to both the previous MCTS version of SyntheMol as well as an AI-based virtual screening baseline. We synthesized 79 SyntheMol-RL compounds that were unique relative to the training dataset and found that 13 showed potent in vitro activity, of which seven were structurally novel after detailed literature searches. Furthermore, one hit compound, synthecin, demonstrated efficacy in a murine wound infection model of methicillin-resistant S. aureus (MRSA). These results validate SyntheMol-RL’s ability to generate novel and synthetically accessible candidate antibiotics and position SyntheMol-RL as a powerful tool for drug design across therapeutic domains.

Year:

2025

Foundational models have emerged as powerful tools for addressing various tasks in clinical settings. However, their potential development to breast ultrasound analysis remains untapped. In this paper, we present BUSGen, the first foundational generative model specifically designed for breast ultrasound image analysis. Pretrained on over 3.5 million breast ultrasound images, BUSGen has acquired extensive knowledge of breast structures, pathological features, and clinical variations. With few-shot adaptation, BUSGen can generate repositories of realistic and informative task-specific data, facilitating the development of models for a wide range of downstream tasks. Extensive experiments highlight BUSGen's exceptional adaptability, significantly exceeding real-data-trained foundational models in breast cancer screening, diagnosis, and prognosis. In breast cancer early diagnosis, our approach outperformed all board-certified radiologists (n=9), achieving an average sensitivity improvement of 16.5% (P-value<0.0001). Additionally, we characterized the scaling effect of using generated data which was as effective as the collected real-world data for training diagnostic models. Moreover, extensive experiments demonstrated that our approach improved the generalization ability of downstream models. Importantly, BUSGen protected patient privacy by enabling fully de-identified data sharing, making progress forward in secure medical data utilization. An online demo of BUSGen is available at https://aibus.bio.

Year:

2025

BACKGROUND: Recent large language models (LLMs) have demonstrated significant advancements, particularly in their ability to serve as agents thereby surpassing their traditional role as chatbots. These agents can leverage their planning and tool utilization capabilities to address tasks specified at a high level. This suggests new potential to reduce the burden of administrative tasks and address current healthcare staff shortages. However, a standardized dataset to benchmark the agent capabilities of LLMs in medical applications is currently lacking, making the evaluation of LLMs on complex tasks in interactive healthcare environments challenging. METHODS: To address this gap to the deployment of agentic AI in healthcare, we introduce MedAgentBench, a broad evaluation suite designed to assess the agent capabilities of large language models within medical records contexts. MedAgentBench encompasses 300 patient-specific clinically-derived tasks from 10 categories written by human physicians, realistic profiles of 100 patients with over 700,000 data elements, a FHIR-compliant interactive environment, and an accompanying codebase. The environment uses the standard APIs and communication infrastructure used in modern EMR systems, so it can be easily migrated into live EMR systems. RESULTS: MedAgentBench presents an unsaturated agent-oriented benchmark that current state-of-the-art LLMs exhibit some ability to succeed at. The best model (Claude 3.5 Sonnet v2) achieves a success rate of 69.67%. However, there is still substantial room for improvement which gives the community a next direction to optimize. Furthermore, there is significant variation in performance across task categories. CONCLUSION: Agent-based task frameworks and benchmarks are the necessary next step to advance the potential and capabilities for effectively improving and integrating AI systems into clinical workflows. MedAgentBench establishes this and is publicly available at https://github.com/stanfordmlgroup/MedAgentBench, offering a valuable framework for model developers to track progress and drive continuous improvements in the agent capabilities of large language models within the medical domain.

Year:

2025

Sleep is a fundamental biological process with broad implications for physical and mental health, yet its complex relationship with disease remains poorly understood. Polysomnog-raphy (PSG), the gold standard for sleep analysis, captures rich physiological signals but remains underutilized due to challenges in standardization, generalizability, and multimodal integration. To address these limitations, we developed SleepFM, a multimodal sleep founda-tion model trained with a novel contrastive learning approach that accommodates multiple PSG montages—the specific arrangements of electrodes and sensors used to record physi-ological signals during sleep. Trained on a curated dataset of over 585,000 hours of PSG recordings from approximately 65,000 participants across multiple cohorts, SleepFM produces latent sleep representations that capture the physiological and temporal structure of sleep and enable accurate prediction of future disease risk. SleepFM achieved a C-Index of at least 0.75 (Bonferroni-corrected p < 0.01) for 130 conditions, including all-cause mortality (C-Index: 0.84), dementia (0.85), myocardial infarction (0.81), heart failure (0.80), chronic kidney disease (0.79), stroke (0.78), and atrial fibrillation (0.78). Moreover, the model demonstrates strong transfer learning performance on a dataset from the Sleep Heart Health Study (SHHS), a dataset that was excluded from pretraining, and performs competitively with specialized sleep-staging models such as U-Sleep and YASA on common sleep analysis tasks, achieving mean F1 scores of 0.70–0.78 for sleep staging and accuracies of 0.69 and 0.87 for classifying sleep apnea severity and presence. This work shows that foundation models can extract clinically meaningful features from multi-modal sleep recordings, enabling scalable, label-efficient analysis and disease prediction.

Year:

2025

Year:

2024

Single-cell sequencing has revolutionized our understanding of cellular heterogeneity and responses to environmental stimuli. However, mapping transcriptomic changes across diverse cell types in response to various stimuli and elucidating underlying disease mechanisms remains challenging. Studies involv- ing physical stimuli, such as radiotherapy, or chemical stimuli, like drug testing, demand labor-intensive experimentation, often hindering the rapid advancement of mechanistic insight and drug discovery. To address this, we present Squidiff, a diffusion model-based generative framework designed to predict transcriptomic changes across diverse cell types in response to a wide range of environmental changes. We demonstrate Squidiff’s robustness across various scenarios, including cell differentiation, gene perturbation, and drug response prediction. Through continuous denoising and semantic feature integration, Squidiff effectively learns transient cell states and predicts high-resolution transcriptomic landscapes over time and conditions. Furthermore, we applied Squidiff to model the development of blood vessel organoids and cellular responses to neutron irradiation and growth factors. Our results demonstrate that Squidiff enables in silico screening of cell molecular landscapes, facilitating rapid hypothesis generation and providing valuable insights for precision medicine.

Year:

2024

Liquid biopsies have the potential to revolutionize cancer care through non-invasive early detection of tumors. Developing a robust liquid biopsy test requires collecting high-dimensional data from a large number of blood samples across heterogeneous groups of patients. We propose that the generative capability of variational auto-encoders enables learning a robust and generalizable signature of blood-based biomarkers. In this study, we analyze orphan non-coding RNAs (oncRNAs) from serum samples of 1050 individuals diagnosed with non-small cell lung cancer (NSCLC) at various stages, as well as sex-, age-, and BMI-matched controls. We demonstrate that our multi-task generative AI model, Orion, surpasses commonly used methods in both overall performance and generalizability to held-out datasets. Orion achieves an overall sensitivity of 94% (95% CI: 87%–98%) at 87% (95% CI: 81%–93%) specificity for cancer detection across all stages, outperforming the sensitivity of other methods on held-out validation datasets by more than ~ 30%.

Year:

2024

Year:

2024

Year:

2024

Year:

2024

Year:

2024

Year:

2024

Year:

2024

Year:

2024

Year:

2024

Collaborators (25)

Olivier Elemento

Professor

Weill Cornell Medical College

UNITED STATES
View details

David Chen

Assistant Professor

Washington University in Saint Louis School of Medicine

UNITED STATES
View details

Pierre Elias

Columbia University Irving Medical Center

UNITED STATES
View details

Yeh-Hsing Lao

Assistant Professor

University at Buffalo

UNITED STATES
View details

Aaron Newman

Associate Professor

Stanford University

UNITED STATES
View details

Ulrike Peters

Fred Hutchinson Cancer Research Center

UNITED STATES
View details

Elham Azizi

Irving Associate Professor

Columbia University

UNITED STATES
View details

Fereydoun Hormozdiari

Associate Professor

University of California

UNITED STATES
View details

Zhenqin Wu

Assistant Professor

The University of Hong Kong

HONG KONG
View details

Emmanuel Mignot

Stanford University School of Medicine

UNITED STATES
View details

Faisal Mahmood

Associate Professor

Harvard Medical School

UNITED STATES
View details

Aadel A. Chaudhuri

Washington University School of Medicine

UNITED STATES
View details

M. Brandon Westover

Stanford University

UNITED STATES
View details

Hongyu Zhao

Professor

Yale University

UNITED STATES
View details

Gordana Vunjak-Novakovic

University Professor

Columbia University

UNITED STATES
View details

Ryan C. Fields

University of Rochester

UNITED STATES
View details

Hani Goodarzi

Associate Professor

University of California, San Francisco

UNITED STATES
View details

Raju Tomer

Assistant Professor

Columbia University

UNITED STATES
View details

Mohammed Abuul Usmani

Senior Scientist

Washington University in St. Louis

UNITED STATES
View details

Eric D. Brown

Professor

McMaster University

CANADA
View details

Kam Leong

Professor

Columbia University

UNITED STATES
View details

Christopher Li

Fred Hutchinson Cancer Research Center

UNITED STATES
View details

Gregory W Charville

Associate Professor

Stanford University

UNITED STATES
View details

Andreas Brink-Kjær

Technical University of Denmark

DENMARK
View details

Eran Segal

Professor

Weizmann Institute of Science

ISRAEL
View details