Bayesian Inference of gene-miRNA regulatory networks

Nowadays, in the post-genomics era, one of the major tasks and challenges is to decipher how genes are regulated. The miRNAs play an essential regulatory role in both plants and animals. It has been estimated that about 30% of the genes in the human genome are down-regulated by microRNAs (miRNAs), short RNA molecules which repress the translation of proteins of mRNAs in animals and plants. Genes which are regulated by a miRNA are called targets of this given miRNA. Hence, the task is to try to determine which miRNAs regulate which genes, in order then to build a network of these DNA components. Knowledge of the functional miRNAs-genes interactions can help find the source or reason of a genetic disease, then we can focus on drugs and their effects such we get more efficient treatments. In this thesis, we aim to build a Bayesian graphical model that infers a regulatory network by integrating miRNAs expression levels with their potential mRNA targets. We incorporate biological information, such as structure and sequence information, via the prior probability model. The method is broken down to 3 stages. First, a dimensionality reduction is performed; the gene expressions are narrowed down by using biological information (association scores and type of probe set), and distance similarity procedures such as clustering of correlated or co-expressed variables. Second, a Bayesian graphical model is proposed, according to which associations of gene and miRNA expressions are inferred, and an association matrix is extracted. The methodology uses simulation-based methods, as Markov Chain Monte Carlo, and benefits by managing uncertainty at a complex network. Finally, using the association matrix, the regulatory network is constructed