CAREER: Machine Learning with Rich Data Sources and Interrelated Tasks

University Of Wisconsin-Madison, Madison WI

Investigators

Abstract

Now that numerous genomes have been sequenced, a significant challenge confronting biologists is to determine the functions of the genes contained in these genomes. One aspect of understanding the function of a given gene is to determine the conditions under which it is active, the mechanisms responsible for controlling its level of activity, and the interactions it has with other genes. Toward this end, in the current project the PI will develop new computational approaches to uncovering the regulatory mechanisms and interactions of genes in a given organism. In particular, the focus is on developing new machine learning methods which are able to predictively identify various regulatory elements of a genome, using well-characterized aspects of the genome as training data. The expected impact of this research is twofold: it will produce new methods and software that can be applied by molecular biologists to gain insight into the regulatory apparatus of the cell; and it will advance the state of the art in machine learning by developing new methods for problem domains that involve (i) multiple inter-related learning tasks, (ii) rich and varied sources of data including sequence and text data, and (iii) the need for rich representation languages, such as stochastic context-free grammars and relational rules.