Doctoral Proposal Defense
Bryan Hall, Room 509c
Strategies for Increasing the Applicability of Biological Network Inference
Advisor: Dr. Michael Brent
One of the fundamental goals of Systems Biology is to understand how networks of genes interact to regulate the transcription response to external stimuli. In the last decade, many researchers have developed algorithms aimed at inferring the structure and kinetics of regulatory networks. These algorithms attempt to learn a network that best explains the measurements of all RNA transcripts in cells grown under various experimental conditions (RNA profiles). A common approach to learning network structures is to find, for each gene ga, another gene or group of genes whose RNA level best explains ga's RNA level. In addition to network structure, the most recent algorithms learn a dynamical system to model the network, by inferring the strength of each regulator's influence on the expression of each of its target genes. Learning these strength-of-regulation parameters allows the algorithms to perform quantitative modeling in which the abundance of each RNA (network state) can be predicted under novel experimental conditions. These algorithms have aided in the study and understanding of several biological systems. However, they have achieved only moderate structural accuracy and poor quantitative accuracy. To increase the accuracy and relevance of these algorithms I will address several shortcomings of common approaches.
In this work, I propose to develop a more accurate algorithm for inferring the structure and dynamics of transcriptional regulatory networks. To accomplish this goal I plan to incorporate known transcription factor binding affinities into a network inference algorithm as a priori knowledge to guide the selection of networks to more accurate results. In addition, I plan to infer the effects of post-transcriptional regulation by treating transcription factor activity levels as latent variables in an expectation-maximization procedure. Also, I plan to relate network genes to the phenotypes they affect by selecting genes whose RNA measurements best explain the quantitative measurements of each phenotype. This will allow us to create more accurate structural and steady state quantitative models of many systems. I will evaluate the algorithm's structural accuracy by inferring the gene regulatory network relating sugar feeding to transcriptional response in a fly model of type 2 diabetes. Quantitative accuracy will be evaluated by comparing network state predictions against unseen network state measurements. Finally, I will deploy the algorithm as a web application allowing experimentalists to search for modifications of regulator expression levels that are predicted to result in a desired phenotype.
Computer Science & Engineering
Back to Calendar