Drug targets prioritization for neglected diseases

SANTIAGO VIDELA; FERNAN AGUERO; ARIEL CHERNOMORETZ

Congreso; IV International Society for Computational Biology-Latin America Bioinformatics Conference; 2016

International Society for Computational Biology

Background: Over the past decade, several methods have been proposed to aid the drug discovery using complex networks. On the one hand, nodes in the network may describe genes, chemical compounds, diseases, tissues, or biological annotations. On the other hand, edges between nodes capture their interactions or relationships. Notably, complex networks allow us to integrate large-scale and highly heterogeneous datasets in a unified framework. Next, the challenge is how to query such heterogeneous networks in order to find valuable insights. Importantly, successful approaches could help to reduce drastically the cost and time required by traditional drug development.Results: We made use of available data from model and non-model organisms to identify candidate drug targets in neglected pathogen genomes. Towards that end, we built a heterogeneous network modeling 1.4 x 106 chemical compounds, the complete genomes of 8 model and 29 pathogens species plus almost 200 incomplete genomes accounting for 1.6 x 105 proteins, and 3 types of biological annotations, namely, Pfam domains, orthology groups, and metabolic pathways. Two types of edges are considered between chemical compounds, namely, Tanimoto similarity and molecular weight symmetry among substructure relationships. Further, known bioactivities are modeled by edges between the corresponding chemical compound and its targets or species (when specific targets are unknown). Further, genes are connected through biological annotations whenever they share   Pfam domain, an ortholog group, or they participate in the same metabolic pathway. We have implemented the described network using a popular graph database, namely, Neo4j. Next, we have adapted various prioritization methods from the recommender systems literature. Using such methods we address two challenging problems. Firstly, given a species of interest we prioritize its complete genome according to the likelihood of being a drug target. Secondly, for a given chemical compound with known bioactivity on certain species but with unknown specific target, we prioritize the complete genome of such species according to the likelihood of being the compound´s target. We show the success of our integrative network approach by cross-validation and early-retrieval performance metrics.Conclusions: Our complex networks approach allows us to integrate a vast amount of highlyheterogeneous information about chemical compounds, their known bioactivities across many species, and various biological annotations of interest. Further, by using a graph database like Neo4J we are able to query this network in a very intuitive and efficient way. In particular, we have implemented various prioritization methods to address two specific problems related to the prioritization of putative drug targets. Broadly speaking, our work shows the advantages of adopting complex networks and graph databases in the field of integrative systems biology. More specifically, we contribute to the subject of drug targets priotitization in neglected diseases, where due to the lack of information, the integration of large-scale and heterogeneous datasets is crucial to find valuable insights and to generatenew hypotheses.