Bioinformatics Seminar

Friday , April 3, 4:00pm, Lecture Hall (room 135), Golden LEAF Biomanufacturing Training and Education Center
Speaker:	Dr. Charles Lawrence, Brown University
Title:	Abuse of the Mode in Genomics and an Ensemble Alternative: Forgotten Role of Entropy
Abstract:	For decades mode based estimators have dominated prediction problems in genomics and computational molecular biology including: structure prediction by minimizing the free energy, sequence alignment via maximum similarity, maximum a-posteriori (MAP) identification of transcription factor binding sites ,and maximum likelihood estimates of ancestral states. However, recent findings in all these areas challenge this paradigm. Advances in genomics have rendered increasingly large data sets available for analysis. While the emergence of such large data sets would seem to lead to increasingly more precise estimates of parameters, paradoxically just the opposite seems to becoming increasingly common. This paradoxical circumstance has emerged because these technologies have simultaneously opened opportunities to draw inferences on previously unanswerable high dimensional questions. Through the lens of RNA secondary structure prediction we examine the untoward effects of over reliance on the mode based point estimates, and present alternative “ensemble based” centroid estimators. We review findings showing that centroid predictions make 30% fewer mistakes in predicting the base pairs of RNA structures while also correctly predicting more base pairs, improve protein structural prediction by homology, and the prediction of transcription factor binding sites. We employ statistical decision theory to examine the properties of centroid estimators, to illustrate that similar results are expected in a broad spectrum of problems, and show how Bayesian confidence limits can be used to asses the reliability of point estimates in discrete high-D spaces.

Page last updated: March 22, 2005

Home NCSU Home Contact Site Map E-mail Webmaster