Extracción, cálculo de similitudes y análisis de características biológicas para la creación de redes humanas complejas

Abstract

The increasing availability of biological data is improving the understanding of diseases and providing new insight into their underlying relationships. A recent hypothesis claims that two diseases that share genes, symptoms or other molecular mechanisms, could have unknown relationships. The connection between pairs of diseases can be used for drug repositioning by checking whether they have biological relationships, and if the drug used to treat any of the diseases could affect some of the biological traits in the relationship. To evaluate these connections and their potential use for drug repositioning, complex human disease networks based on molecular information have been created. Relationships between diseases in complex human disease networks can be formed using different biological traits, such as symptoms, genes, or protein-protein interactions (PPIs). Thanks to these complex human disease networks, a greater understanding of the functioning of the diseases will be accomplished, and the analysis of these networks with computational approaches will allow this understanding to reach a larger scale. The DISNET project, in which is involved the current project, aims to build a comprehensive complex human disease network in three levels: the phenotypical layer, the genetical or biological layer and the drugs layer. The current project pretends to construct the DISNET biological layer programmatically in the most possible automatized way. Therefore, the state of the art of the current biological public databases is analyzed, in order to access to them and extract the necessary information to build the network. Genes, proteins, metabolic pathways and PPIs are selected to be the biological traits to create it. The process to access the resources and obtain the data is developed by Python scripts. Genes and proteins are queried to DisGeNET, metabolic pathways are queried to WikiPathways and PPIs are queried to IntAct. This data is structured in a relational way and stored in the same database of the DISNET phenotypical layer. A method to compute the similarities between diseases and the analysis of these similarities is also included. Disease similarity is represented by three metrics: cosine similarity, Jaccard index and Dice coefficient, which are computed for each combination of two diseases without repetition and based in each trait. The obtained values of these metrics are compared to the ones obtained in the phenotypical layer. In future lines, machine learning techniques will be applied to these results to obtain new information from the data.