Drug Repurposing for Rare Diseases

This repository contains the code for an analytical pipeline that utilizes data from open-source chemical databases to identify novel compound-gene target relationships for rare disease drug prioritization. More specifically, the pipeline includes clustering of compounds from the Toxicology in the 21st Century (Tox21) program using SOM, identifying compound-gene target relationships using data from NIH's Pharos and Broad Institute's Drug Repurposing Hub, and performing gene enrichment analysis to obtain a list of enriched gene targets for each cluster.

Key markdown files

• clustering.Rmd: Cleans the original Tox21 dataset tox21_data.txt and uses the kohonen R package to perform SOM clustering.

• compound_gene_relationships.Rmd: Identify compound-gene relationships for the 7170 compounds (note: 8971 original compounds --> data cleaning & keeping only compounds with complete bioassay data --> 7,170 compounds) using the databases Pharos and Drug Repurposing Hub. In the end, we have 7,030 compounds with the chemical structure identifier SMILES (6503 from parent smiles list and 527 from the aggregated tox21 data).

• gene_enrichment_analysis.Rmd: Uses results from compound_gene_relationships.Rmd to identify a list of enriched gene targets for each cluster. The Bum class is used to address the issue of multiple comparisons.

Key data files

• tox21_data.txt: The original data file for the entire project; contains the bioassay activity profile data for each of the Tox21 compounds. A list of bioassays used in the Tox21 program can be found here. Compound activity was measured by curve rank, a value between -9 and 9 determined by the potency, efficacy and quality of the concentration-response curve (a large positive number represents strong activation and a large negative number represents strong inhibition of the assay target). The file contains 8,971 compounds and 243 variables.

• tox21_aggregated.txt: Additional aggregated chemical information for the Tox21 compounds; variables of interest include the CAS RN number, sample name, assay outcomes(i.e., active, inactive, inconclusive), SMILES, TOX21_ID, and etc.

• hgnc_data.txt: Contains the HGNC ID, approved gene symbol, and approved gene name for all the gene targets. Custom downloads could be obtained here. This dataset is used to better understand what each gene target represents (e.g., A1BG = "alpha-1-B glycoprotein", ACACA = "acetyl-CoA carboxylase alpha").

• tox21_10k_cas_smiles_without_salts.txt: Contains the CAS Registry Number and the parent SMILES for a selection of the Tox21 compounds.

• broad_reformatted.txt: Data pulled and organized from the Broad Institut's Drug Rep Hub. The data version used is from 3/24/2020 and can be obtained here. Variables include CompoundName, TargetGene, PubchemCID, InKey, MOA, and etc.

• pharos_all_target_ligands-p1/2.csv: Data from NIH's Pharos database; can be obtained here. Variables of interest include UniProt, Symbol, Ligand Name, Ligand SMILES, and etc. Note that the data file is split in two parts due to its large file size.

• rdkit Subfolder: The files under this subfolder are obtained by running RDKit in Python. Briefly, the open-source cheminformatics tool RDKit is used to find the corresponding inchikeys for the SMILES identifier.

  • smi_to_inchikey.csv: Contains the cas, cluster, smiles, and inchikey for all of the Tox21 compounds.
  • pharos_to_inchikey.csv: Contains the smiles, symbol, and inchikey for all compounds pulled from Pharos.

Key result files

• clusters.txt: Contains the results from SOM clustering (i.e., which cluster does each compound belongs to); generated from clustering.Rmd.

• enrichment_analysis_results.csv: Generated from gene_enrichment_analysis.Rmd. Contains a list of gene targets and their approved gene names for each cluster. A column on whether the gene target is considered significantly enriched in the cluster is also included (obtained by using fisher's exact test and then keeping a overall false discovery rate (FDR) of 1% by applying a beta-uniform distribution).

• data_w_identifers.csv: Generated from gene_enrichment_analysis.Rmd. Contains more identifiers for each compound, including CAS, pubchem id, sample name, smiles, inchikey, gene targets (symbol and approved name).

• cluster_and_smiles.csv: Contains the tox21 compounds,the cluster they belongs to, and their parent smiles. This is the final dataset used in Rdkit python code to get the corresponding inchikey for the Tox21 compounds.

• pharos_and_symbols.csv: Contains compound-gene target associations pulled from Pharos. This is the dataset used in Rdkit python code to get the corresponding inchikey for Pharos compounds. Column headers: smiles, symbol.

• inchikey Subfolder: The files under this subfolder are the interim result files obtained from compound_gene_relationships.Rmd, by mapping the Tox21 compounds with Pharos/Drug Repurposing Hub compound-target list, using inchikey as the primary key.

  • drug_hub_inchikey.csv: Compound and gene target pairs obtained from Broad institute's drug repurposing hub. The column headers are: cas, cluster, inchikey, short_key (i.e, the first 14 characters of the inchikey), and target_gene.

  • pharos_inchikey.csv: Compound and gene target pairs obtained from NIH's Pharos.

  • pharos_and_drughub.csv: Contains unique compound-gene target pairs obtained after combining the drug repurposing and the Pharos dataset. Combined, this datafile contains 1829 unique compounds and 1629 unique target genes. This is the in put data file for gene_enrichment_analysis.Rmd.