Biostatistics seminar

About

Lunch will be provided during the seminar.

Generative AI for Drug Discovery: From Statistical Foundations to Real-World Impact

This talk will be led by Xia Ning, professor of biomedical informatics, computer science and engineering, and pharmacy at Ohio State.

Artificial intelligence (AI) in drug discovery has advanced far beyond predicting the properties of existing molecules. Recent advances in generative AI (genAI) have opened unprecedented opportunities to accelerate drug discovery and enable molecular design in ways that were previously unimaginable. However, the statistical foundations underlying these advances are often overlooked, making generative AI appear almost magical rather than grounded in rigorous mathematical and statistical principles.

In this talk, I will present our research on generative AI for drug discovery, focusing on two applications: (1) de novo drug design, in which generative models create new small-molecule drug candidates—including their atomic compositions, chemical bonds, and three-dimensional geometries—conditioned on protein binding pockets (i.e., structure-based drug design); and (2) scaffold hopping, in which generative models design structurally novel molecules while preserving desired functional groups.

Through these examples, I will uncover the statistical principles underlying modern generative AI, illustrate how these principles shape the design and capabilities of generative models, and demonstrate their potential to advance real-world drug discovery.

Prioritized Composite Outcomes and Win-Based Methods: Concepts, Extensions, and Applications

This talk will be led by Rejuan Haque, research assistant professor in the department of biomedical informatics at Ohio State.

Clinically important outcomes such as mortality or severe adverse events often occur too infrequently for a single primary outcome to provide adequate statistical power. Composite outcomes address this by combining several clinically relevant endpoints, but conventional analyses typically weight all components equally, despite their differing clinical importance, or focus on time to first event, ignoring subsequent events and other meaningful benefits. Prioritized composite outcomes analyzed with win-based methods overcome these limitations by building the relative importance of each outcome directly into an interpretable framework for comparing treatments.

In this presentation, I will introduce the fundamental concepts of prioritized composite outcomes. I will discuss key methodological considerations and advantages over conventional approaches. I will then highlight my methodological research that extends win-based methods through distance-based weighting. To illustrate their practical relevance, I will discuss applications in suicide prevention research, where treatment effectiveness involves multiple outcomes of varying clinical importance, and introduce ongoing research under the WINDSURFER project, where we will be using hierarchical composite outcomes. Finally, I will discuss the broader opportunities and challenges associated with win-based methods, including outcome prioritization, clinical interpretability, sensitivity to endpoint definitions, and methodological limitations.

Contact

Andy Ni

---