Empirical revision of biological networks using machine learning. (360G-Wellcome-080712_Z_06_Z)

Empirical Revision of Biological Networks using Machine Learning It is well known that existing public domain biological network descriptions contain topological errors of omission and commission. This PhD project aims to develop and evaluate methods to effectively and efficiently revise the knowledge deposited in public biological networks (like Kyoto Encyclopedia of Genes and Genomes) by using machine learning techniques. We plan to start the theory revision system by using facts and background knowledge from well characterised networks such as the aromatic amino acid pathways of yeast. We will gradually remove randomly selected information using existing theory revision systems to reconstruct the deleted portions of the system. At some point it is expected that the amount of information deleted will be so significant that full recovery is impossible. Doing this systematically for a different group of metabolic networks known organisms will allow us to characterise the limitations of the available techniques for biological data. Having assessed the limitations of existing techniques we will develop methods based on the new area of Statistical Relational Learning to overcome these limitations.