|
Abstract:
One of the current challenges of molecular biology is to understand the function of genes encoded in the human genome, and discover what mechanisms regulate their expression. For practical considerations, this information is typically inferred by comparing the human genome to that of close relatives such as the mouse. However, we need to be sure the gene has maintained its function for the inference to be correct. Some recent studies on simpler organisms have had success using multispecies gene expression measurements in order to infer points along the evolutionary tree at which changes in gene function have likely occurred. In this talk, I will present a variant of a mixture of factor analyzers that describe where in a tree changes in gene expression patterns have occurred, given parallel gene expression datasets in multiple organisms. This model represents a significant expansion of previously defined models, as the constrained factor loading matrices allow consideration of the relationship between tissues in a given organism, and therefore can handle the considerably higher complexity of species such as humans. I will also present our results on training this model on a novel gene expression dataset collected by our collaborators from the human, mouse, chicken, frog, and tetraodon species. This work is done in collaboration with Yee Whye Teh, Esther Chan, Michael Brudno, and Timothy Hughes.