Synthetic genetic interactions reveal buffering mechanisms in the cell against genetic perturbations. These interactions have been widely used by researchers to predict functional similarity of gene pairs. In this talk, I present results from a comprehensive evaluation of various methods for predicting co-pathway membership of genes based on their neighborhood similarity in the genetic network. We clearly delineate the scope of these methods and use it to motivate a rigorous statistical framework for quantifying the contribution of each pathway to the functional similarity of gene pairs. We then use our model to infer interdependencies among KEGG pathways. The resulting KEGG crosstalk map yields significant insights into the high-level organization of the genetic network and is used to explain the effective scope of genetic interactions for predicting co-pathway membership of gene pairs. A direct byproduct of this effort is that we are able to identify subsets of genes in each pathway that act as `ports' for interaction across pathways.
More generally, in the talk, I will also highlight various statistical and algorithmic approaches used in biological network analysis.
Various parts of this talk draw on work with Shahin Mohammadi, Giorgos Kollias, Mehmet Koyuturk, Shankar Subramaniam, and Wojciech Szpankowski.