Home > Program

 

Detailed Program

Registration

Tutorial 1

AI Applications in Pharma
Workshop Coordinators: Nareen Katta, AbbVie; Subha Madhavan, Pfizer Inc.
Claire Zhao, Pfizer; Nunzio Camerlingo, Pfizer; Haoda Fu, Amgen

Break

Tutorial 2

Targeted Learning
Workshop Coordinators: Nareen Katta, AbbVie; Subha Madhavan, Pfizer Inc.
Mark van der Laan, University of California, Berkeley; Susan Gruber, Targeted ML Solutions; and Hana Lee, US Food and Drug Administration

Registration and Coffee

Welcome

Kannan Natarajan, Pfizer, and Ron Wasserstein, American Statistical Association

Opening Remarks

Special Guest: Benjamin E. Hermalin, University of California, Berkeley
Moderator: Mark van der Laan, University of California, Berkeley

Keynote

Personalized Medicine and Treatment Optimization
Robert Tibshirani, Stanford University
Moderator: Elizabeth Stuart, Johns Hopkins University

Plenary Session 1

Chair: Eren Demirhan, Genentech

From Imaging to Insight: AI Applications in Drug Discovery and Clinical Trials
Ziad Obermeyer, University of California, Berkeley
Medical images and waveforms are central to medical practice but have been largely absent from research. AI tools, by engaging directly with detailed physiological information in images, open up a vast new set of improvements in the way we discover and test drugs, from clinical trial enrichment to principled surrogate outcomes. I’ll illustrate the potential of these methods with a few examples from my own work.


Realizing Value from AI in Drug Development: Rigorous Integration of AI Prediction Models in Clinical Trials
Christina Rabe, Genentech
While generative AI for workflow optimization and automation tasks dominate the current discussion, realizing the value of clinical AI models remains a significant challenge. Traditional machine learning often suffices for tabular data; however, the shift toward high-dimensional, less structured inputs, such as medical imaging, presents an untapped opportunity to enhance trial efficiency.

In this talk, we focus on the rigorous application of prognostic models within randomized clinical trials (RCTs). Models that are prognostic for trial outcomes based on baseline variables are often used for enrichment, stratified randomization, imbalance checking, and subgroup analyses. However, it is not widely recognized that one of the most powerful uses of prognostic baseline variables in randomized clinical trials is for covariate adjustment. We provide practical guidance on building, evaluating, and deploying these models specifically for this use case. We propose a framework to quantify a model's potential benefit measured by Effective Sample Size Increase (ESSI) prior to trial execution. We demonstrate this approach using a deep-learning model trained on ophthalmology imaging data. Our results show that such models can be integrated into primary endpoint analyses to substantially increase precision and power over simple baseline variables. We align these methods with recent FDA guidance on covariate adjustment and AI. By adhering to pre-specified statistical principles, sponsors can safely leverage advanced AI to improve decision-making and power without compromising the integrity of treatment effect estimation. Finally, we will point out interesting areas of statistical research to further advance this field and conclude by discussing the risks and regulatory implications of other potential prognostic use cases in the clinical landscape.


Harvesting Quick Wins: Practical AI in Pharmaceutical Medicine
Chenguang Wang, Regeneron
AI—especially large language models—is showing up in nearly every corner of pharmaceutical medicine. Yet the exact path by which it will transform end-to-end drug development remains unclear. While frontier applications such as discovering novel genetic targets and multimodal biomarkers may require scientific breakthroughs, immediate value is available in low-complexity, high-volume tasks: Systematically automating the mundane can unlock disproportionate gains in speed, quality, and cost. In this talk, we outline a two-dimensional strategy for harvesting quick wins—practical AI deployments that improve everyday workflows without overhauling core science or governance—and examine the future of the pharmaceutical biometrics function in the AI era. "

Break & Poster Session

Poster Coordinators: Javier Cabrera, Rutgers University, and Larry Han, Northeastern University

Plenary Session 2

Chair: Mike Daniels, University of Florida

Generative AI Enables Extracellular Vesicle-Based Immunotherapy
Kiley Graim, University of Florida

Immunotherapy can prime the patient's immune system against tumor cells, significantly improving anti-tumor immunity while eliminating chemotherapy-associated toxicity. While AI models have accelerated drug discovery and development, a significant barrier remains in designing effective targeting peptides for immunotherapeutic applications. To address this need, we have developed a generative AI framework, ExoGAN, that designs peptides that stimulate a desired immune response on a common cancer biomarker, MHC-I. ExoGAN is a generative adversarial network (GAN) that integrates HLA physiochemistry feature engineering with sequence-level data. It uses quantum computing to calculate chemical and physical features of peptides to contextualize physiochemical interaction dynamics. Trained on the largest IEDB data set of HLA peptides, ExoGAN designs neoantigens that bind specific MHC-Is (HLAs) with improved TCR recognition and T cell activation. For effective peptide delivery, we incorporate membrane surface molecular engineering to decorate extracellular vesicles (EVs) with ExoGAN-designed peptides in MHC complexes. Computational and experimental validation demonstrate that ExoGAN-generated peptide-carrying EVs consistently present antigens to T cell TCRs, resulting in precisely controlled and programmable agents for anti-tumor immunity activation. This approach substantially accelerates therapeutic development timelines and mechanistic understanding of neoantigen functionality despite variable TCR signaling across patients. It demonstrates the clinical feasibility of AI-designed targeting peptides.


Catalyze Next-Generation Drug Innovation with Imaging AI
Yalei Chen, Pfizer
The rapid advancement of deep learning and artificial intelligence has fundamentally transformed biomedical image analysis, driving the evolution of imaging AI from narrowly task-specific algorithms to more generalizable foundation models capable of learning rich, multi-scale representations across diverse modalities. Imaging has long remained a cornerstone of how humans observe, interpret, and understand biological systems, and it continues to play a critical role in clinical decision-making across the drug discovery and development continuum—from target identification and phenotypic screening to translational biomarker and novel clinical endpoints development. Recent innovations in imaging AI have made it increasingly feasible to integrate advanced computational methods into day-to-day biomedical workflows. These advances enable scalable, quantitative, and reproducible analysis of complex imaging data generated in early discovery, preclinical research, and late-stage clinical development. By augmenting human expertise with AI-driven image analysis, imaging AI holds significant potential to accelerate drug development timelines and improve decision quality. Beyond gains in operational efficiency, adoption of AI in imaging analysis enables novel biological insights, supporting more informed target selection, mechanism-of-action characterization, and patient stratification. With the growth of curated AI-ready data and advancements in imaging AI, its thoughtful integration into drug discovery and development pipelines is poised to accelerate, improve, and innovate therapeutic development.


Latent Embedding Alignment for Sample-Efficient Brain Encoding and Decoding
Lexin Li, University of California, Berkeley
Brain encoding and decoding aims to understand the relationship between external stimuli and brain activities and is a fundamental problem in neuroscience. We study latent embedding alignment for brain encoding and decoding, with a focus on improving sample efficiency under limited fMRI-stimulus paired data and substantial subject heterogeneity. We propose a lightweight alignment framework equipped with two statistical learning components: inverse semi-supervised learning that leverages abundant unpaired stimulus embeddings through inverse mapping and residual debiasing and meta transfer learning that borrows strength from pretrained models across subjects via sparse aggregation and residual correction. Both methods operate exclusively at the alignment stage while keeping encoders and decoders frozen, allowing for efficient computation, modular deployment, and rigorous theoretical analysis. We establish finite-sample generalization bounds and safety guarantees and demonstrate competitive empirical performance on the large-scale fMRI-image reconstruction benchmark data.

Lunch & Poster Session

Poster Coordinators: Javier Cabrera, Rutgers University, and Larry Han, Northeastern University

Keynote

Clinical Study Decision-Making in the AI Era
Narimon Honarpour, Amgen
Moderator: Amy Xia, Amgen

Plenary Session 3

Chair: Haoda Fu, Amgen

Targeted Machine Learning to Improve Global Health
Laura B. Balzer, University of California, Berkeley
We address common challenges arising in randomized trials and RWE studies in global health. Specifically, we highlight the potential for bias and misleading inference due to missing data and dependent data. We showcase how the Causal Roadmap and Targeted Machine Learning Estimation (TMLE) provide a robust framework for addressing these and other challenges. We illustrate with real-world examples from the SEARCH consortium to prevent HIV and improve global health.


AI Teammates in Clinical Development: Integrating Clinical Insights with Statistical Innovations
Will Ma, Hope AI
While Large Language Models excel at general knowledge, clinical development demands a more nuanced approach that combines domain expertise with statistical rigor. We present a novel dual-agent system consisting of an AI Clinician that synthesizes complex clinical evidence and identifies optimal patient populations and benchmarks, working in tandem with an AI Statistician that conducts meta-analyses and implements real-time statistical simulation for optimal study design. This core partnership is supported by additional specialized agents including AI Regulatory and AI Medical Writers, creating a comprehensive ecosystem of AI teammates. Each agent operates within well-defined domains using curated, structured data sources to ensure precision and auditability. This AI-augmented approach shows significant promise for accelerating and strengthening clinical trial design, enabling evidence-based decision-making while scaling intelligence rather than headcount.


From Policy to Practice: Structural Barriers to Implementing FDA’s Risk-Based Framework for AI in Drug Development
Tala Fakhouri, Parexel
AI is now deployed across the GxP spectrum, spanning regulated manufacturing and clinical processes through to the generation of evidence directly supporting regulatory decision-making in a marketing dossier. Apart from purely operational efficiency applications, this entire landscape falls within the scope of FDA’s 2025 draft guidance, which establishes a risk-based credibility assessment framework calibrated to context of use, model influence, and decision consequence. The conceptual architecture is sound, but the implementation can be challenging.

Three structural barriers impede meaningful operationalization of the guidance. First, organizations frequently misapply computerized system validation approaches to AI models, a legacy approach misaligned to how these systems function and to the risk profiles that credibility assessment is designed to address. Second, organizations routinely adopt “human-in-the-loop” as a self-declared governance solution without specifying which humans hold decision authority, under what criteria they act, and what accountability structures ensure meaningful oversight rather than nominal review. Third, effective implementation demands simultaneous fluency across regulatory science, statistical methodology, and clinical domain knowledge, a combination rarely concentrated within a single organization, compounded by fragmented data access and the absence of cross-sector implementation standards.

This talk examines what meaningful operationalization of FDA’s credibility assessment framework across the GxP spectrum might look like and what industry, regulators, and research partners must collectively build to close the implementation gap."

Break & Poster Session

Poster Coordinators: Javier Cabrera, Rutgers University, and Larry Han, Northeastern University

Panel: Agentic AI to Accelerate Drug Development

Moderator: David Madigan, Northeastern University
Xun Chen, AbbVie; Peter J. Embí, Vanderbilt University Medical Center; Liviu Niculescu, CMO, US Medical Affairs at Novartis; Maya Petersen, University of California, Berkeley; Romain Pirracchio, University of California, San Francisco; and Hoifung Poon, Microsoft

Closing Remarks and Acknowledgments

Demissie Alemayehu, Pfizer

Networking Reception

Special Guest: Jennifer Tour Chayes, University of California, Berkeley