toxicity-cancer-drug-combination

Preprocess the Databases

1. Run test_preprocessing_functions.ipynb and make sure all tests pass. This will ensure that the following have been pre-processed:

• DrugBank Drug-Drug Interactions Database
  • All NA values for drug names and toxicity severity have been removed
  • Stored at data_processed/drugbank_ddi.csv
• DDInter Drug-Drug Interactions Database
  • All NA values for drug names and toxicity severity have been removed
  • Stored at data_processed/ddinter_data.csv
• DrugComb Drug Combination Synergy Scores Database
  • All remaining combinations have all synergy scores (Bliss, HSA, Loewe, S_max, S_mean, S_sum, and ZIP)
  • All remaining combinations have both drugs in the pair named
  • All remaining combinations are limited to cancer drug combinations
  • Stored at data_processed/drugcomb_data.csv
• DrugBank Drug Target Database
  • Parse XML file for drugs to their drug targets, compile into dataframe
  • Stored at data_processed/drugbank_drug_targets.csv
• Reactome Pathway Database
  • Parse Reactome files for lowest pathways and all pathways
  • Lowest pathways stored at data_processed/reactome_lowest_pathways_homo_sapiens.csv
  • All pathways stored at data_processed/reactome_all_pathways_homo_sapiens.csv
• Jaccard Similarity method is working
• Jonckheere Terpestra Test is working
• Known Toxicity and Synergy Score INtersection between DrugComb and DrugBank
  • Stored at data_processed/drugbank_syntox_known.csv
• Known Toxicity and Synergy Score Intersection between DrugComb and DDInter
  • Stored at data_processed/ddinter_syntox_known.csv

2. Find the intersection between all databases

• Run ddinter_parse_toxsyntarpathstring_intersection.ipynb
  • Manually adds in some missing DrugBank Target information, stores all drug targets from the (known DDInter + DrugComb + DrugBank Target) intersection in data_processed/ddinter_syntoxtarg.csv
  • Creates mapping of target proteins to UniProtIDs for mapping to Reactome Pathways/STRING, stored in ddinter_syntoxtarg_UniprotIDs.txt
  • Finds intersection with Reactome lowest and all pathways, stores all drug targets from known DDInter + DrugComb + DrugBank Target + Reactome Lowest/All Pathways intersection in data_processed/ddinter_syntoxtarg_lowestpw.csv and data_processed/ddinter_syntoxtarg_allpw.csv respectively
  • Then finds intersection by mapping UniProtIDs to STRING IDs, storing all drug targets from DDInter + DrugComb + DrugBank Target + Reactome Lowest/All Pathways + STRING in data_processed/ddinter_syntoxtarg_lowestpw_string.csv and data_processed/ddinter_syntoxtarg_allpw_string.csv respectively
  • Then finds the drug combinations that contain all information between DDInter + DrugComb + DrugBank Target + Reactome Lowest/All Pathways + STRING with the average target distance, stored in data_processed/ddinter_processed_combos_syntoxtargallpw_string.csv
  • Note, did not have to additionally filter by SMILES/Morgan Fingerprint representations because found all of these for DDInter
• Run drugbank_parse_toxsyntarpathstring_intersection.ipynb
  • Adds in some missing DrugBank Target information, from the academic licensed version of DrugBank, stores all drug targets from the (known DDInter + DrugComb + DrugBank Target) intersection in data_processed/drugbank_syntoxtarg.csv
  • Creates mapping of target proteins to UniProtIDs for mapping to Reactome Pathways/STRING, stored in drugbank_syntoxtarg_UniprotIDs.txt
  • Finds intersection with Reactome lowest and all pathways, stores all drug targets from known DDInter + DrugComb + DrugBank Target + Reactome Lowest/All Pathways intersection in data_processed/drugbank_syntoxtarg_lowestpw.csv and data_processed/drugbank_syntoxtarg_allpw.csv respectively
  • Then finds intersection by mapping UniProtIDs to STRING IDs, storing all drug targets from DDInter + DrugComb + DrugBank Target + Reactome Lowest/All Pathways + STRING in data_processed/drugbank_syntoxtarg_lowestpw_string.csv and data_processed/drugbank_syntoxtarg_allpw_string.csv respectively
  • Then finds the drug combinations that contain all information between DDInter + DrugComb + DrugBank Target + Reactome Lowest/All Pathways + STRING with the average target distance, stored in data_processed/drugbank_processed_combos_syntoxtargallpw_string.csv
  • Note, did not have filter by SMILES/Morgan Fingerprint representations because found all of these for DDInter

Synergy Score and Toxicity Analysis: Create Figure 2

The first step of the analysis is to examine the relationship between synergy scores and toxicity. To do this and create the panels used in Figure 2 as well as Supplemental Figure S1, there are two files that need to be run:

1. DrugBank: drugbank_syntox_analysis.ipynb
2. DDInter: ddinter_syntox_analysis.ipynb

The figure files are stored in results/figure2/ and the statistics are stored in:

• results/synergy_score_tox_categ_distrib/kruskal_dunn_jonck_majmodmin_syntox_drugbank.tsv
• results/synergy_score_tox_categ_distrib/kruskal_dunn_jonck_majmodmin_syntox_ddinter.tsv

There are supplementary histograms and statistics stored in results/synergy_score_tox_categ_distrib and results/synergy_score_distrib

Drug Target and Pathway Analysis: Create Figure 3 and Figure 4

The second step of the analysis is to examine which drug targets and pathways contribute to synergy scores and toxicity. To create Figure 3, Figure 4, and Supplemental Figure S2, the following files should be run:

1. DrugBank: drugbank_targpw_analysis.ipynb
2. DDInter: ddinter_targpw_analysis.ipynb

The figure files and statistical test results are stored in results/jaccard_similarity_analysis/

Toxicity Score Analysis: Create Figure 5

Figure 5 has three major components:

1. Assessing whether the average distance of targets on the STRING PPIN trends with toxicity levels -- run:

• ddinter_avg_string_distance_analysis.ipynb
• drugbank_avg_string_distance_analysis.ipynb

2. Assessing whether overlapping drug structure trends with toxicity levels -- run:

• ddinter_deeptrasynergy_toxscore.ipynb
• drugbank_deeptrasynergy_toxscore.ipynb

3. Assessing whether overlapping drug target protein neighborhoods on the STRING PPIN trends with toxicity levels -- run:

• ddinter_graphsynergy_toxscore.ipynb
• drugbank_graphsynergy_toxscore.ipynb