Developing, extending, and combining computational methods for the inference of genome-wide genetic regulatory networks

Living cells continually face a diverse set of challenges, for example maintenance of homeostasis, migration between distinct locations, differentiation, and apoptosis. To face these challenges living cells employ gene regulatory networks (GRNs) that are responsible for processing information and carrying out an appropriate cellular response. Characterizing GRNs is a key problem in current biology with applications spanning medicine, bioengineering, and other biological fields.;High-throughput technologies are rapidly developing and producing large heterogeneous datasets that were unimaginable just a decade ago. These developments subsequently drive the rapid development of methods that aim to study GRNs on the genome-scale. Often these methods are designed (and are thus optimal) for learning GRNs from a single type of high-throughput data (e.g. from time-series experiments measured by microarrays). Because each high-throughput data-type provides incomplete information regarding the biological system studied, and since each GRN inference method has its own set of strengths and weaknesses, it is desirable to combine methods into GRN inference pipelines.;In this thesis I describe and use four GRN inference methods, two of which are novel, one of which is an extension to a previously published method, and the last is an already well-established method. Each of these methods has different characteristics with respect to the input data they are optimal for, the type of GRN models they produce, and the types of statistical approaches they employ, and are, for these reasons, complementary. Here, I demonstrate how these individual GRN inference methods can be combined in a mutually reinforcing manner to produce state-of-the-art network inference pipelines. Importantly, the principles and heuristics that I use to combine GRN inference methods are not unique to the method used, and can be generalized to combine other methods as well. Two of the GRN inference pipelines presented in this thesis were top performers in the two recent Dialogue for Reverse Engineering Assessments and Methods (DREAM3-2008 and DREAN4-2009) 100-gene in-silico network inference challenges (ranking 1st and 2nd out 18 and 22 contesting methods, respectively).