Identifying Drug Repositioning Candidates using Comparative Mechanism Enrichment in Neurodegenerative Diseases
Daniel Domingo-Fernández1,2, Charles Tapley Hoyt1,2, Mohammed Asif Emon1,2, Reagon Karki1,2, Martin Hofmann-Apitius1,2
1Department of Bioinformatics, Fraunhofer SCAI, Sankt Augustin 53754, Germany
2Rheinische Friedrich Friedrich-Wilhelms-Universität Bonn, Bonn-Aachen International Center for IT, Bonn 53115, Germany
Classic chemoinformatics approaches for predicting structure activity relationship (SAR) focus on the relationships between small molecules and their protein targets. However, the investigations of single targets are often insufficient for understanding complex disease in which the modulation of either a constellation of targets or entire mechanisms (i.e., pathways) is likely necessary. Dissimilar molecules whose targets appear consecutively in a given mechanism could have similar effects in phenotypic screens that can not be explained with SAR.
Unraveling this complex biology requires the usage of prior knowledge of mechanisms that incorporate relationships between proteins that are directional, causal, and polar. The mechanism enrichment algorithm published by Domingo-Fernández et al. (2017) can be used to support this kind of reasoning in networks arising from knowledge assemblies encoded in Biological Expression Language (BEL). Though it was originally used to interpret gene lists arising from high throughput genomics experiments, we adapted it to handle drugs and drug-like chemicals on the basis of their targets (e.g., from DrugBank or ChEMBL) in order to assess their perturbations of the candidate disease mechanisms for Alzheimer’s disease (AD) and epilepsy in NeuroMMSig. We presented a case study in which we proposed that the disease-specific perturbations of the GABA-ergic receptor signaling pathway could explain the unknown mechanism of action through which the drug Carbamazepine has been seen to have positive clinical results for both AD and epilepsy patients (Hoyt et al. 2018).
Now, we have applied this methodology systematically for all mechanisms in all diseases listed in NeuroMMSig (i.e. AD, Parkinson’s disease, and epilepsy) and for all drugs listed in DrugBank that have entered at least Phase 1 clinical trials. We first looked to validate the methodology on a larger scale by assessing that previously approved drugs did in fact enrich the NeuroMMSig mechanisms with their nominal targets. Next, we used a guilt-by-association approach to propose candidate drug-disease associations for diseases with mechanisms similar to mechanisms enriched for the same drug for another diseases. Finally, we prioritized the proposed drug-disease pairs using an orthogonal mechanism enrichment workflow based that uses real-world data from OpenTargets, GWAS-DB, GWAS Catalog, GRASP, and PHEWAS data sets to evaluate if a drug could revert the pathological differential gene expression profiles of its proposed associated disease.
We will soon package these workflows as a reusable Python packages that can accept any disease-specific mechanisms encoded in BEL (or related systems biology modeling languages) and be interchanged with other evaluation methods.
References
[1] Domingo-Fernández, D. et al. (2017). Multimodal Mechanistic Signatures for Neurodegenerative Diseases (NeuroMMSig): a web server for mechanism enrichment. Bioinformatics, 33(22), 3679-3681.
[2] Hoyt, C. T., et al. (2018). A systematic approach for identifying shared mechanisms in epilepsy and its comorbidities. Database, 2018.
