Statistical tools for classification, interpretation and prediction of biological data

• Autores: Oriol Senan Campos
• Directores de la Tesis: Marta Sales Pardo (dir. tes.), Roger Guimerà Manrique (codir. tes.)
• Lectura: En la Universitat Rovira i Virgili ( España ) en 2017
• Idioma: español
• Tribunal Calificador de la Tesis: Xavier Correig Blanchar (presid.), M. Angeles Serrano Moral (secret.), Jaeyun Sung (voc.)
• Materias:
  • Química
    • Bioquímica
• Enlaces
  • Tesis en acceso abierto en:  hdl.handle.net  TDX 
• Resumen

  • Mathematical modeling have been used for more than 100 years in theoretical biology, but now has become a fundamental part of biological research.

    The improvement of omics technologies is making possible the massive profiling of DNA, RNA, proteins and metabolites in living organisms. This data opens the possibility for a paradigm shift towards a systemic approach. Despite the sequentiation of the genome and a wide coverage of the proteome, one fundamental question remains open: How many different metabolites are in a given organisms or in a biological sample? With the current best technique for detecting the maximum number of metabolites, a liquid chromatography coupled to mass spectrometry (LC/MS), oftenly only 20-30 metabolites are identified over the thousands of signals in the data. CliqueMS, a novel method for grouping and annotating the multiple signals produced per metabolite in LC/MS. This multiple signals are produced by isotopic variants, different ionizations of the metabolite, called adducts, and fragmentations.

    To improve this annotation we have developed. Our method outperforms current annotating methods and may contribute to overcome one of the main bottlenecks for a better annotation of metabolomic experiments.

    The first application we expect with our new method is to compute the distribution of adducts in untargeted metabolomics experiments. The estimation of this distribution will be a prior information that will improve further annotation of new untargeted metaboloic experiments. Despite this and other inconvenients of omics technologies, there is a constant growth in the recorded biological data. We need a combination of the multiple sources of biological data to achieve a better comprehension of the system as a whole. This combination won’t be straightforward, as we do not see simple associations, for example there is no general correlation between mRNA and proteins abundance. Integration of data demand new mathematical models, to unveil the complex relations between the different biomolecules.

    How to integrate all this data? Mathematical models are validated by its capacity for prediction, but are good predicting models also good for interpretation? In the second paper wee evaluate the role of different Computational models for predicting platelet deposition. Platelet deposition is the trigger of thrombus formation, a very important pathology. In this study we analyze the interplay between prediction and interpretability of the different models, and the importance of several variables onto platelet deposition. We demonstrate that by measuring platelet concentration, vessel tissue and other input variables of our models, we can predict the platelet deposition in a new sample. We expect that a better approximation to thrombus formation will be to integrate the spatial information of platelet deposition, and additionally to include the effect of fibrinogen in our model.

    Finally, regarding the combination of multiple omics data, we study the therapeutiacal effect of Hibbiscus Sabdariffa extracts in humans by analyzing the metabolomic and transcriptomic response after its ingestion. In the third and last paper included in this thesis we first report the molecular composition of Hibbiscus Sabdariffa extracts. We observe an alteration of the immune response, the mitochondrial function and the energy homeostasis, revealed by the patterns of expression and the metabolic responses.