Beatriz García Jiménez

"Functional Annotation of Proteins based on Relational Representatiton in the Systems Biology Context." June 4. 2012. Universidad Carlos III de Madrid. [PDF] Supervised by Alfonso Valencia and Araceli Sanchis.
National Best Thesis Award by the Royal Spanish Doctoral Academy (Experimental and Technological Sciences, 2012)
Show abstractHide abstract

Functional annotation is an open and interesting research topic in Molecular Biology. Determining a function in terminology terms is a hard task, due to lack of unified criterion and also because a function takes up many levels for the same protein. Given this difficulties, the way to determine a protein function is to annotate it with several terms from different vocabularies. Proteins carry out their function together with other proteins, being part of protein complexes. These interactions are represented in a network of experimentally verified protein-protein interactions. Analyzing and using the interaction network is task of interest due to the great number of associations, and to the multiple ways in which a protein could influence in the function of others. Therefore, this thesis focuses in the prediction of functional annotation based on networks. It's apparent that this complex scenario couldn't be faced without computational techniques. In fact, in Computational Biology, there is a considerable activity specially devoted to this topic. This thesis is part of this effort for applying computational methods to biological problems in the Systems Biology area. This approximation can belong to the Systems Biology context, because it does not analyze function in an isolated way for each molecule, but at system level, taking into account all the relations among genes and proteins linked at different levels. To take advantages of all these biological relations, and to preserve their structured semantics, this thesis suggests to use Relational Representation, since in particular it is suitable for the concerning domain. Over such representation, multiple transformations and Artificial Intelligence techniques are applied to retrieve implicit knowledge from the related proteins, and to propose new functions through the prediction of functional associations between proteins. The main proposal of this thesis is to characterize the function of proteins and genes based on networks, through Relational Representation and Machine Learning. Specially, from a relational representation specific to functional annotation, we look for the computational design needed to solve two specific, biological interesting and different problems. The former consists of predicting functional association between pair of proteins in E.coli, and the latter comprises expanding pathways in humans. We perform an assessment in computational and biological interpretation terms. Besides, we propose new putative protein functional annotations to be experimentally verified. In addition, the thesis investigates diverse approaches to knowledge representation and learning techniques, suggesting specific strategies to tackle other biological problems, specially where relational data or multi-class and multi-label targets are present.

Media appearance Interview (in Spanish)