Analysis of the PC algorithm as a tool for the inference of gene regulatory networks: evaluation of the performance, modification and application to selected case studies.

The expansion of a Gene Regulatory Network (GRN) by finding additional causally-related genes, is of great importance for our knowledge of biological systems and therefore relevant for its biomedical and biotechnological applications. Aim of the thesis work is the development and evaluation of a bioinformatic method for GRN expansion. The method, named PC-IM, is based on the PC algorithm that discovers causal relationships starting from purely observational data. PC-IM adopts an iterative approach that overcomes the limitations of previous applications of PC to GRN discovery. PC-IM takes in input the prior knowledge of a GRN (represented by nodes and re- lationships) and gene expression data. The output is a list of genes which expands the known GRN. Each gene in the list is ranked depending on the frequency it appears causally relevant, normalized to the number of times it was possible to find it. Since each frequency value is associated with precision and sensitivity values calculated using the prior knowledge of the GRN, the method provides in output those genes that are above the value of frequency that optimize precision and sensitivity (cut-off frequency). In order to investigate the characteristics and the performances of PC-IM, in this thesis work several parameters have been evaluated such as the influence of the type and size of input gene expression data, of the number of iterations and of the type of GRN. A comparative analysis of PC-IM versus another recent expansion method (GENIES) has been also performed. Finally, PC-IM has been applied to expand two real GRNs of the model plant Arabidopsis thaliana.