Seeking best practices in classifier construction and testing

“Seeking best practices in classifier construction and testing”

Tuesday, September 30, 2014 at 12pm

Dr. Clark Jeffries, University of North Carolina

This webinar featured Dr. Clark D. Jeffries, a Bioinformatics Scientist and Adjunct Professor at the University of North Carolina at Chapel Hill, presenting the topic “Seeking Best Practices in Classifier Construction and Testing.”


At UNC-Chapel Hill’s Eshelman School of Pharmacy, Dr. Jeffries’ research focuses on the design and interpretation of assays of small RNAs as agents of signaling that may be causally upstream of many other types of biomarkers. In the webinar, Dr. Jeffries will discuss the following challenge: while the construction and testing of classifiers of cases versus controls is a fundamental task of research programs, widespread failure to replicate findings remains a concern for research sponsors. One side of replication is testing the conclusions of a classifier with external data or random relabeling of samples. Many methods can be used to generate classifier functions from data, but the relative utility of these methods remains debatable, and modern, computer-based methods still find seemingly convincing patterns in random data. Hence, rigorous testing is needed, often in the absence of external data. To this end, Dr. Jeffries presented findings on a robust classifier method and a stringent testing method.