A stimulant (like a virus) to a cell can interact with proteins in the cell and cause a cascade of signaling pathways that activate/repress the expression of genes downstream. The differential expression of such genes can cause further signaling cascades that cause the differential expression of another set of genes. While the gene expression at different time points can be observed, it is not possible to directly observe the signaling pathways responsible for their differential expression. We developed a tool, TimePath, to learn the progression of such signaling pathways across time based on time series gene expression data and protein-protein interaction data. The paper is currently under submission.
We developed a tool, MT-SDREM, to jointly learn signaling pathways and regulatory networks using time series gene expression data from multiple related conditions, TF-gene interaction data, and a protein-protein signaling network. Our tool built on an existing tool called SDREM which learnt the signaling pathways and regulatory network for just a single condition. For joint learning, we shared information the following information :- (1) we ensured that the condition-specific networks learnt were consistent – i.e. the edge directions were the same for all networks (2) If a TF was predicted to regulate more than one condition, it’s prior for regulating any condition was increased. Validation with respect to RNAi screen hits, GO analysis and manual examination of the predicted signaling proteins demonstrated the advantage of joint inference. Read more about it here.
We programmed a Mixed Integer programming solver from scratch (using existing LP solvers) and experimented with various node selection and branching heuristics with application to the Warehouse Location problem.