Research Vision
I place equal weight on developing new statistical methodology and making it usable in the scientific settings that motivate it. Both are central to my research: advancing statistical ideas and translating those advances into tools that directly improve how scientists design experiments, analyze data, and draw conclusions.
Statistics has long developed through a close relationship with practice. Scientific challenges generate fundamental statistical questions, while statistical advances help other disciplines build more rigorous approaches to evidence. I see this exchange as essential to the field’s future. Engagement with applications shapes the questions worth asking, the assumptions worth studying, and the standards by which a method should be evaluated.
Rapid advances in AI give this relationship renewed urgency. AI tools can help researchers without specialized statistical training carry out increasingly sophisticated analyses. This creates valuable opportunities, but producing an analysis and establishing its statistical soundness remain distinct challenges. An analysis can run successfully and yield convincing results even when its assumptions are inappropriate or its uncertainty is poorly quantified. As the ability to perform complex analyses expands, we need methods and tools that make statistically justified decisions equally accessible.
My research responds to this need by pursuing methodological advances with direct consequences for scientific work. In perturbation genomics, this means addressing calibration, power, and computation together: controlling false discoveries, helping scientists design informative experiments, and making rigorous inference feasible at the scale of modern data. The scientific setting informs both the theoretical questions and the practical requirements of the methods I develop.
I view software as an integral part of this research. My goal is to deliver tools that domain scientists can use in their own workflows without custom modifications or ad hoc statistical tuning. This requires principled defaults, useful diagnostics, and clear communication of assumptions and limitations. Making a method accessible means carrying its statistical principles through to the decisions users make, so that methodological progress becomes part of everyday scientific practice.
