Bioinformatics applied to human genomics and proteomics: development of algorithms and methods for the discovery of molecular signatures derived from omic data and for the construction of co-expression and interaction networks

Abstract

[EN] The present PhD dissertation develops and applies Bioinformatic methods and tools to address key current problems in the analysis of human omic data. This PhD has been organised by main objectives into four different chapters focused on: (i) development of an algorithm for the analysis of changes and heterogeneity in large-scale omic data; (ii) development of a method for non-parametric feature selection; (iii) integration and analysis of human protein-protein interaction networks and (iv) integration and analysis of human co-expression networks derived from tissue expression data and evolutionary profiles of proteins.

In the first chapter, we developed and tested a new robust algorithm in R, called DECO, for the discovery of subgroups of features and samples within large-scale omic datasets, exploring all feature differences possible heterogeneity, through the integration of both data dispersion and predictor-response information in a new statistic parameter called h (heterogeneity score). In the second chapter, we present a simple non-parametric statistic to measure the cohesiveness of categorical variables along any quantitative variable, applicable to feature selection in all types of big data sets. In the third chapter, we describe an analysis of the human interactome integrating two global datasets from high-quality proteomics technologies: HuRI (a human protein-protein interaction network generated by a systematic experimental screening based on Yeast-Two-Hybrid technology) and Cell-Atlas (a comprehensive map of subcellular localization of human proteins generated by antibody imaging). This analysis aims to create a framework for the subcellular localization characterization supported by the human protein-protein interactome. In the fourth chapter, we developed a full integration of three high-quality proteome-wide resources (Human Protein Atlas, OMA and TimeTree) to generate a robust human co-expression network across tissues assigning each human protein along the evolutionary timeline. In this way, we investigate how old in evolution and how correlated are the different human proteins, and we place all them in a common interaction network.

As main general comment, all the work presented in this PhD uses and develops a wide variety of bioinformatic and statistical tools for the analysis, integration and enlighten of molecular signatures and biological networks using human omic data. Most of this data corresponds to sample cohorts generated in recent biomedical studies on specific human diseases.