Drug repurposing: databases and pipelines

Abstract

The concept of drug repurposing is focused on the repositioning of drug molecules that have already undergone safety trials. There are different strategies for drug repurposing. Network-based strategy focuses on the evaluation of drug combinations in a molecular environment with multi-target hits and analysis of drug interactions. Implementation of any in silico strategy requires several databases and pipelines for executing the process of shortlisting appropriate drugs.

Adherence to the notion of one drug used for one protein as treatment has failed to provide productive solutions for several disorders caused by multiple diverse factors. Currently, pharmaceutical companies are aiming to reposition approved drugs as treatment for various diseases as de novo drug designing is not only slow, but also cumbersome.

The concept of drug repurposing is focused on the sensible repositioning of a plethora of drug molecules that have already undergone safety trials thereby avoiding extensive investment. Drug repurposing is evaluated as a low-risk and highly efficient strategy due to the investment of $1.6 billion, which is nearly one-tenth of the monetary commitment required for de novo designing of a drug, ¹ antibiotic resistance, ² rare diseases, ³ oncological studies, ⁴ inflammatory disorders, ⁵ and neurological diseases ⁶ highlights its importance.

The scope of drug repurposing broadened massively since the COVID-19 pandemic and most of the drugs used to treat moderate to severe COVID-19 infections were initially screened using repurposing pipelines. ⁷

Different approaches like signature matching, genetic association, and pathway mapping can be used for drug repurposing. ⁸ Two main strategies are namely drug-based, wherein a known FDA-approved drug is chosen as the starting point and either on-target or off-target drug repositioning is performed, and target-based, wherein a target molecule or biomarker is chosen as the starting and all FDA-approved drugs are screened against this target molecule using computational approaches. ⁹

Network-based approach facilitates the evaluation of drug combinations in a molecular environment with multi-target hits and analysis of drug interactions. ¹⁰ One of the key strategies is interactome construction using different genes and related gene products. ¹¹ Networks called diseasomes link multiple disorders and disease-related genes to provide a better insight into how different diseases are linked to each other and this helps pinpoint those genes which form a major link between seemingly unrelated diseases and behave as targets of drug repositioning. ¹¹ Network medicine has proven to play a vital role in drug repurposing in identifying the drug targets by calculating the proximity scores between drug targets and disease genes in the human protein–protein interactome. ¹²

Implementation of any in silico strategy requires extensive research using databases, and pipelines for executing the process of shortlisting appropriate drugs. Every drug repositioning pipeline requires a collection of all the FDA-approved drugs and their pharmacological properties, structural information of drugs and proteins, gene-level analytical resources, knowledge of the disease, and the pathways related to the disease.

There are different categories of databases commonly used in drug repurposing.

Molecular structure of drugs: ChEMBL ¹³ (https://www.ebi.ac.uk/chembl/) is a manually curated database containing drugs and molecules with high bioactivity. PubChem ¹⁴ (https://pubchem.ncbi.nlm.nih.gov/) is a database of chemical molecules by name, structure, molecular formula, biological activities, physical properties, safety, and other identifiers. DrugBank ¹⁵ (https://go.drugbank.com/) is a manually curated comprehensive database of drugs and their targets, groups, pharmacological parameters, interactions, transporters, mechanism of action, and useful properties. ChemSpider ¹⁶ (http://www.chemspider.com/) is a chemical structure database providing all the properties, literature review, vendors, and similar molecules to a particular ligand.

Drug targets: BindingDB ¹⁷ (https://www.bindingdb.org/rwd/bind/index.jsp) provides information on binding affinities and interactions of all ligand molecules with the target and off-target proteins. Drug Target Commons ¹⁸ (https://drugtargetcommons.fimm.fi/) provides annotations regarding bioactivities of drug–target interactions and relation with multiple proteins in the body accompanied by bioassay data. STITCH ¹⁹ (http://stitch.embl.de) produces a detailed map of the drug–protein interactions and further protein–protein relations to identify all possible drug targets. TTD ²⁰ (https://db.idrblab.net/ttd/) is an organized platform for identifying drugs for a specific target protein along with patient data and literature. DrugCentral ²¹ (https://drugcentral.org/) has up-to-date drug information based on protein targets for repositioning.

Drug side effects and toxicity: SIDER ²² (http://sideeffects.embl.de/) is a collection of the drug indications and recorded adverse drug reactions of all marketed drugs. TOXRIC ²³ (https://toxric.bioinforai.tech/home) is a novel repository with metabolic reactions and categorized toxicological data for all drugs.

Protein and Gene related: UniProt ²⁴ (https://www.uniprot.org/) has the highest quality information regarding the structural and functional annotation of discovered proteins. GproteinDB ²⁵ (https://gproteindb.org/?) stores structural, functional, mutational, and endogenous ligand information of all G-coupled proteins in the body. PDB ²⁶ (https://www.rcsb.org/) has 3-D X-ray crystallographic and NMR structures of proteins along with important structural annotations. SCOPe ²⁷ (https://scop.berkeley.edu/) is a highly curated structural classification of proteins focusing on protein similarity and interactions. HuRI ²⁸ (http://interactome-atlas.org/) has mapping of the interaction of the target protein with all other proteins in the body along with their relationships. HINT ²⁹ (http://hint.yulab.org/) stores binary and co-complex protein interactions mapped to generate interactome. STRING ³⁰ (https://string-db.org/) provides all the functional associations of a protein in the form of an interactome along with sequence similarity, disease pathway, and latest annotations. OMIM ³¹ (https://www.omim.org/) is a thorough assembly of all human genes with reference to phenotypic expression in disorders, structure, location, mapping, and functioning. GenBank ³² (https://www.ncbi.nlm.nih.gov/genbank/) is genetic sequence collection with protein translation information and gene similarities. BioGRID ³³ (https://thebiogrid.org/) stores detailed description of the gene–gene and gene-chemical interactions. DGIdb ³⁴ (https://www.dgidb.org/) has curated details on all the gene interactions for a drug and vice versa. IntAct ³⁵ (https://www.ebi.ac.uk/intact/home) provides molecular interactions specifically protein–protein and RNA-protein interactions. GEO ³⁶ (https://www.ncbi.nlm.nih.gov/geo/) accepts array and sequence-based data and provides tools to query and download experiments and curated gene expression profiles.

Pathways: KEGG ³⁷ (https://www.genome.jp/kegg/) stores detailed map of the molecular reactions, relations, and interactions along with the biochemistry of the disease. Reactome ³⁸ (https://reactome.org/) has representation of entire pathways, proteins, reactions, and interactions specific to diseases and protein complexes in a hierarchical manner. Pathway Commons ³⁹ (https://www.pathwaycommons.org/) is used for illustrious pathway visualization for various biochemical processes and diseased states.

Disease-associated: Phenopedia ⁴⁰ (https://phgkb.cdc.gov/PHGKB/startPagePhenoPedia.action) offers summarized information in a disease-centric view on human genetic associations. DisGeNET ⁴¹ (https://www.disgenet.org/) is a large collection of genes and variants associated with human diseases and integrates data from various repositories to provide various gene-disease and variant-disease associations. DISEASES—Database Commons ⁴² (https://diseases.jensenlab.org/) provides records of the major genes, proteins, and experiments relating to a disease.

Repurposed Drugs: ReDO_DB ⁴³ (https://www.anticancerfund.org/en/redo-db) stores information on compounds that are noncancer drugs showing anticancer activity. DrugRepV ⁴⁴ (https://bioinfo.imtech.res.in/manojk/drugrepv/) stores manually curated drugs and chemicals that display antiviral activity against epidemic and pandemic viruses. ReFRAME ⁴⁵ (https://reframedb.org/) contains nearly all small molecules that have reached clinical development or undergone significant preclinical profiling.

Drug repurposing pipelines focus on obtaining the best set of candidate drug molecules for the selected disease through rigorous computational techniques and intensified analytical measures.

Computational Analysis of Novel Drug Opportunities (CANDO) ⁴⁶ (https://github.com/ram-compbio/CANDO) is an open-source platform for the analysis of drug interactions on a proteomic scale. In CANDO drug–protein interaction signatures are generated from an extensive library of known interaction mappings and compared and screened based on their similarities to the drugs used for the same disease. In total, 51 of the 276 molecules predicted by this platform against SARS-CoV-2 are explored in clinical studies and have demonstrated promising activity. ⁴⁷

KsRepo ⁴⁸ (https://github.com/adam-sam-brown/ksRepo) is an R-based open-source expression-level platform for drug repurposing that identifies potential candidates enrichment scores. KsRepo analyses and compares RNA-seq data to known gene–drug interactions and also exhibits flexibility in data set types and can also predict drug candidates with limited information on drug–gene interactions. ⁴⁹

SperoPredictor ⁵⁰ is a generic repurposing framework consisting of multiple machine learning algorithms like Random Forest, Tree Ensemble, and Gradient Boosted Trees to predict repurposable drug candidates. SperoPredictor pipeline involves data collection, training, and subsequent deployment of machine learning models followed by literature and molecular docking-based validation.

Single-cell Guided Pipeline to Aid Repurposing of Drugs (ASGARD) ⁵¹ (https://github.com/lanagarmire/ASGARD) makes use of scRNA-seq data of single cell and cell clusters in the disease to predict potentially repurposable drugs. ASGARD processes scRNA-seq data for differential gene analysis and identifies drug candidates which reverse gene expression.

One of the most recent approaches to drug repurposing is using network medicine to identify biomarkers and target molecules through interactomes. ⁵² Searching off-lAbel dRUg aNd NEtwoRk (SAveRUNNER) ⁵³ (https://github.com/sportingCode/SAveRUNNER) is a R-based tool that predicts drugs that can be repositioned by preparing a drug-disease association with the help of human interactome. SAveRUNNER prepares a drug-disease association by calculating the network-based proximity scores which show how close the drug molecule is to the disease in the human interactome. It has been successfully used to repurpose drugs for diseases such as COVID-19, ⁵⁴ Amyotrophic Lateral Sclerosis, ⁵⁵ Breast cancer, ⁵⁶ and cardiovascular disease. ⁵⁷

The Konstanz-Integration Miner (KNIME) ⁵⁸ (www.knime.com) is an open-source platform used to generate semi-automated workflows which are intensely used for drug repurposing. The approach used in KNIME is based on mining of data from various exhaustive databases and processing, analyzing, and visualizing this data in order to shortlist the drugs that can be repurposed.

Conclusion

With an increasing demand to find cures for diseases using faster and more cost-effective methods, drug repurposing is proving to be the future of medicine and the more sensible approach as compared to de novo drug design by reducing the number of clinical and experimental trials and making strong predictions about which drugs can be repositioned, thus, paving the way for successful treatments.