Machine Learning Methods for Discovery of Regulatory Elements in Bacteria

Опубликовано 6 сентября 2016, 5:18

I will present novel machine learning methods for the discovery of important DNA sequence elements encoded in bacterial genomes. Knowledge of these elements provides insight into important problems in computational biology such as uncovering gene functions, gene-regulatory networks, and evolutionary relationships among genes and organisms. This talk will focus on my contributions related to the design and learning of graphical probability models of these elements. In particular, I will present methods for (i) refining the structure of stochastic context-free grammars, (ii) training sequence models with weakly labeled data, (iii) designing models that incorporate multiple and diverse evidence sources and (iv) modeling and predicting arbitrarily overlapping elements in sequence data. The results of cross-validation experiments on the heavily studied bacterium E. coli show that the accuracy of our predictions exceeds the previous state-of-the-art.