Seminar on Computational Learning and Adaptation

  Incorporating Biological Knowledge into the Evaluation of Causal Regulatory Hypotheses

Lonnie Chrisman
Institute for the Study of Learning and Expertise, and
Center for the Study of Language and Information, Stanford University

Biological data can be scarce and costly to obtain. Many factors including high dimensionality, small sample sizes and unobserved variables typically limit statistical power (the ability to distinguish real effects from spurious ones), making reliable inference of causal relations extremely difficult. One approach to dealing with this problem is to incorporate prior domain knowledge into data analysis. I introduce a framework for testing whether an experimental data set contains statistically significant support for a causal or regulatory relationship in the context of domain background knowledge. Causal hypothesis testing of this type is typically considered beyond the scope of classical statistical hypothesis testing methods. I demonstrate that incorporating domain background knowledge into data analysis can substantially improve statistical power, making it possible to detect real causal relationships at a statistically significant level where the same relationships are indistinguishable from noise from the data alone.

Return to the seminar schedule