Drug target repositioning in neglected tropical diseases using a tripartite network-based approach

BERENSTEIN AJ; MAGARIÑOS MP; AGÜERO F; CHERNOMORETZ A

Cordoba

Workshop; XIII Latin American Workshop on Nonlinear Phenomena; 2013

Background Neglected tropical diseases (NTDs) are human infectious diseases that occur in tropical or subtropical regions and are often associated with poverty. Historically, lack of interest from the pharmaceutical industry, resulted in the lack of good drugs to combat the majority of the pathogens that cause these diseases. Recently, the availability of open chemical information has increased with the advent of public domain chemical resources and the release of data from high throughput screening assays. In our laboratory, our goal is to prioritize and identify candidate drug targets, and candidate drug-like molecules to foster drug development in Trypanosoma cruzi (causative agent of Chagas disease). For this we use comparative genomics, and chemogenomics approaches, taking advantage of the availability of drug-target data from other model organisms that have been extensively studied, like human, yeast, and mouse. Materials and Methods Chemical datasets, including bioactivity data against pathogen and non pathogen targets were obtained from open databases and high throughput screenings. Starting from these data, we built a tripartite network considering three disjoint set of vertexes with approximately 1.7 105 drugs and 1.7 105 proteins across more than 150 species and several biological relations (orthology, sharing of Pfam domains, participation in defined metabolic pathways), organized in three different planes. Three different classes of target similarity criteria were considered: sharing of PFAM domains present in the same protein, clustering in the same ortholog group (OrthoMCL algorithm), and belonging to the same metabolic pathway. Only statistically significant terms (in context of drug-target predictions) were taken into account. A bipartite projection was made using a modified version of the Zhou method [2] over the protein plane. In the resulting monopartite protein-protein network, proteins are linked if and only if, they share at least one relevant biological entity. Finally, in order to get a prioritization list of potential targets, a voting scheme was performed using all known sets of drug-targets associations. Results We performed a cross validation procedure by splitting drug-target evidence in two sets: an evaluation set (target proteins of a given organism) and a training set (all drug-target evidence in the remaining organisms). In preliminary tests, we aimed to retrieve (recall) known/validated E. coli and mouse targets after omitting all links from these species. The information contained in the network (derived from other organisms) allowed us to identify several drug-targets from these with AUCs of ~ 0.89 and 0.73 respectively. These results suggests that is possible to identify candidate drug targets, even in the absence of species-specific protein inhibition data. This is particularly important in the case of neglected diseases, as this provides a mean to leverage data from model organisms (or from other tropical diseases) to guide drug repositioning exercises in an organism/disease of interest.