Abstract

A large number of recent studies wink at the idea that every biological state can be described by a proper gene expression signature: a well defined set of genes together with a pattern of expression linked to it. The underlying concepts are: i) any condition,(e.g. the activity of a given pathway, a disease phenotype or cellular response to drug treatment), induces some change in transcriptional activity; ii) even if a single gene on its own poorly characterizes a biological state of interest, this ability is significantly improved when considering a set of genes with their combined pattern of expression.

In this scenario DNA micro-array technology can be deemed as the natural language through which these “biological-state-summaries” can be generated. Using novel non-parametric genome-wide metrics, novel drug-to-disease and/or drug-to-drug “connections” have been predicted comparing the corresponding gene-expression signatures. Several approaches have been proposed to exploit the massive amount of publicly available gene-expression data for computational “drug re-purposing” (i.e. predicting new therapeutic applications for already existing drugs).

In contrast with these methods (very good predictors but providing limited mechanistic insight), transcriptomic data following drug treatment have been integrated with pathway maps and protein-protein-interaction networks to infer upstreaming molecular events, thereby dissecting the mechanisms involved in drug-response. ! We will describe a number of approaches based on the introduced concepts, highlighting pros and cons of each classes of methods. Finally, we will present some successful case studies and useful public available tools and resources.

The talk is part of the CCBI seminar series and the DTP graduate course Reviews in Computational Biology, but is open to all attendees.