Data-Driven and Error-Controlled Hypothesis Generation for Biomedical Discovery

Rapid advances in high-throughput sequencing and single-cell technologies now enable the collection of large-scale multi-omics datasets at unprecedented resolution. Yet extracting reliable biological insight from these heterogeneous, high-dimensional measurements remains a major bottleneck. As datasets grow, the space of plausible explanations expands even faster, making it increasingly difficult for a purely hypothesis-driven workflow to identify which signals are real and worth validating.

In this talk, I will describe a data-driven paradigm for generating testable hypotheses directly from biomedical data in a way that is both interpretable and statistically trustworthy. I will present a series of model-agnostic interpretation methods that identify important features and non-additive interactions learned by modern machine learning models. Crucially, the resulting features and interactions are treated as candidate hypotheses and are reported with rigorous error control, providing confidence guarantees without relying on classical p-values. This line of work helped establish that machine learning interpretations can be made not only useful, but also reliable enough to guide targeted experimental follow-up.

(This talk is part of the weekly Machine Learning Lunch Meetings (MLLM), held every Tuesday from 12:15 to 1:15 p.m.  Professors from Computer Sciences, Statistics, ECE, the iSchool, and other departments will discuss their latest research in machine learning, covering both theory and applications. This is a great opportunity to network with faculty and fellow researchers, learn about cutting-edge research at our university, and foster new collaborations. For the talk schedule, please visit https://sites.google.com/view/wiscmllm/home. To receive future weekly talk announcements, please subscribe to our UW Google Group at https://groups.google.com/u/1/a/g-groups.wisc.edu/g/mllm.)