Reverse engineering of biological signaling networks via integration of data and knowledge using probabilistic graphical models

Abstract

Motivation <br /> The postulate that biological molecules rather act together in intricate networks, pioneered systems biology and popularized the study on approaches to reconstruct and understand these networks. These networks give an insight of the underlying biological process and diseases involving aberration in these pathways like, cancer and neuro degenerative diseases. These networks can be reconstructed by two different approaches namely, data driven and knowledge driven methods. This leaves a critical question of relying on either of them. Relying completely on data driven approaches brings in the issue of overfitting, whereas, an entirely knowledge driven approach leaves us without acquisition of any new information/knowledge. This thesis presents hybrid approach in terms of integration of high throughput data and biological knowledge to reverse-engineer the structure of biological networks in a probabilistic way and showcases the improvement brought about as a result.<br /> Accomplishments<br /> The current work aims to learn networks from perturbation data. It extends the existing Nested Effects Model (NEMs) for pathway reconstruction in order to use the time course data, allowing the differentiation between direct and indirect effects and resolve feedback loops. The thesis also introduces an approach to learn the signaling network from phenotype data in form of images/movie, widening the scope of NEMs, which was so far limited to gene expression data. Furthermore, the thesis introduces methodologies to integrate knoowledge from different existing sources as probabilistic prior that improved the reconstruction accuracy of the network and could make it biologically more rational. These methods were finally integrated and for reverse engineering of more accurate and realistic networks. <br /> Conclusion<br /> The thesis added three dimensions to existing scope of network reverse engineering specially Nested Effects Models in terms of use of time course data, phenotype data and finally the incorporation of prior biological knowledge from multiple sources. The approaches developed demonstrate their application to understand signaling in stem cells and cell division and breast cancer. Furthermore the integrative approach shows the reconstruction of AMPK/EGFR pathway that is used to identify potential drug targets in lung cancer which were also validated experimentally, meeting one of the desired goals in systems biology.