This folder holds the data files, scripts, and results of an analysis
of several publicly available potential drug-drug interaction (PDDI)
knowledge bases. The general format of the project is as follows:

analysis-results -- the data output from analyzing the overlap of the
                    individual datasets
bibliography  --  scientific literature or documents importnat for the analysis
json-data  -- some of the PDDI data was stored in JSON
log-files --  script logs ran to support the analysis
PDDI Datasets -- the PDDI data pulled from the original sources
pickle-data -- PDDI data from each source stored as a Python pickle
scripts -- the code used to query or load the PDDI data int the
           pickles. Also code used for the analysis
Sql scripts -- the sql code used to query, load or map the PDDI data in Relational DB         
terminology-mappings -- mappings of various PDDI data elements to DrugBank
                     -- the database dump of the relational DB used for mappings is saved under here.
                     -- As for the RDBMS, we used Sql server 2012. 
Merged Dataset Location; 
          --Less conservative dataset is saved under; https://code.google.com/p/swat-4-med-safety/source/browse/trunk/SW-DDI-catalog/drug-drug-interactions/analysis-results/CombinedDatasetNotConservative.csv.zip
          --More conservative dataset is saved under https://code.google.com/p/swat-4-med-safety/source/browse/trunk/SW-DDI-catalog/drug-drug-interactions/analysis-results/CombinedDatasetConservative.csv.zip


Please include this citation if you plan to use the merged dataset:

Ayvaz S, Horn J, Hassanzadeh O, Zhu Q, Stan J, Tatonetti NP, et al. Toward a complete dataset of drug–drug interaction information from publicly available sources. Journal of Biomedical Informatics, Volume 55, June 2015, Pages 206-217, ISSN 1532-0464, https://doi.org/10.1016/j.jbi.2015.04.006.

------------------------------
VALIDATION STEPS
------------------------------

We validated both the DrugBank mappings and the conversion of PDDIs
from the sources to the Python PDDI model by.....

the log of the validation is stored in the file; /analysis-results/Validation-List.csv

the manual validation was conducted by selecting the first, last and a few randomly selected records 
from the middle of each overlap result files. Then, the URIs for each drug and the Interaction were used to 
locate the resource in their source to verify the validity of the mapping.
All of the DDIs in the validation file is manually confirmed to be accurate.



------------------------------
DRUGBANK PDDIs
------------------------------


Scanned for duplicates as follows:

<example>

common = {}

for i in range(0,len(pddiL)):
    k1 = "%s-%s" % (pddiL[i]["drug1"], pddiL[i]["drug2"])        
    k2 = "%s-%s" % (pddiL[i]["drug2"], pddiL[i]["drug1"])
    
    if common.get(k1) == None and common.get(k2) == None:
        common[k1] = ""
    else:
        print "dup: %s" % (k1 + ":" + k2)


</example>

------------------------------
NDF-RT PDDIs
------------------------------


NDF-RT to drugbank mappings based on a query against the LinkedSPLs SQL database:

SELECT DISTINCT f.PreferredSubstance, f.Drugbank_Bio2rdf, n.NUI
FROM RXNORM_NDFRT_INGRED_Table n
INNER JOIN FDAPreferredSubstanceToRxNORM a ON n.RxNorm = a.RxNorm
INNER JOIN FDA_SUBSTANCE_TO_DRUGBANK_BIO2RDF f ON f.PreferredSubstance = a.PreferredSubstance
ORDER BY f.PreferredSubstance, n.NUI

The results of the SQL query dumped into the csv file located at ;
	/terminology-mappings/Ndfrt-Drugbank-Mapping.csv

Using the python script "load-NDFRT-Drugbank-Xref-fromCSV.py" the mappings in the csv file converted into 
a pickle file to be used in the overlap analyses. 
The Xref pickle file is stored here; ../pickle-data/ndfrt-ddis-xref.pickle


------------------------------
DIKB PDDIs
------------------------------

Scanned for duplicates as follows:

  for i in range(0,len(allDIKBL)):
        k1 = "%s-%s" % (allDIKBL[i]["drug1"], allDIKBL[i]["drug2"])        
        k2 = "%s-%s" % (allDIKBL[i]["drug2"], allDIKBL[i]["drug1"])
       
        if not common.has_key(k1) and not common.has_key(k2):
            common[k1] = ""
        else:
            print "dup:  %s %s %s" % (k1,k2,allDIKBL[i]["uri"])
           # print allDIKBL[i]
     
------------------------------
TWOSIDES PDDIs
------------------------------

TWOSIDES drugs were mapped to DrugBank as follows:

1) The script query-pubchem-to-drugbank-mapping.py was used to query
DrugBank (http://drugbank.bio2rdf.org/sparql) for pubchem ids. Data
was output to pubchem-to-drugbank-mapping.pickle

2) The script load-TwoSides-DDIs.py loaded the mapping file and the
TSV file with TWOSIDES data. It then iterated through all of the PDDIs
adding bio2rdf namespaces to the pubchem ids so that they could be
matched with drugbank CUIs.

------------------------------
CREDIBLEMEDS.ORG PDDIS
------------------------------

The HTML source from
<http://www.crediblemeds.org/healthcare-providers/drug-drug-interaction>
was downloaded and the div containing PDDIs copied to a text file
(crediblemeds-raw-PDDIs-09282013.txt). Rich Boyce manually copied the
object and precipitant drugs to a spreadsheet. Statins were expanded

by querying LinkedSPL for drug products tagged as NDF-RT class
"HMG-COA REDUCTASE INHIBITOR [EPC]" and then manually copying out the
active moiety strings. NSAIDs were expanded using the list provided by
the FDA in
<http://www.fda.gov/downloads/Drugs/DrugSafety/ucm089162.pdf>. All
drugs were then mapped to drugbank using the reference drugbank
mapping file for linkedSPLs
<https://swat-4-med-safety.googlecode.com/svn/trunk/linkedSPLs/ChEBI-DrugBank-bio2rdf-mapping/fda-substance-preferred-name-to-drugbank-09292012.csv>
and manual searches of
<http://www.drugbank.ca/drugs/DB01015>. SULFAMETHOXAZOLE/TRIMETHOPRIM
was mapped to the drugbank id for SULFAMETHOXAZOLE. THYROID USP was
dropped because that is a preparation of other thyroid hormones and
there is no specific drugbank id. 

------------------------------
KEGG PDDIS
------------------------------

The script query-DIKB-DDIs.py retrieves all DDIs from the
KEGG Medicus API (http://www.kegg.jp/kegg/rest/keggapi2.html). The
script reads all Kegg drugs from all-kegg-drugs-7212014.txt which
pulled from the Kegg FTP site (7/21/2014). The API query is very
simple (e.g., try http://rest.kegg.jp/ddi/D00564) and returns a
tab-delimited text where each line is the Kegg drugs involved in the
interaction, the severity (precaution "P" or contraindiation "C"), and
mechanism (if available). The script parses this data then keeps only
thos interactions for which both drugs can be mapped to DrugBank. The
DrugBank mapping came from a query of Bio2RDF for all drugs with an
xref containing the text "kegg" and is stored in
drugbank-to-kegg-mapping.csv. The Query used to pull the results for the file drugbank-to-kegg-mapping.csv:

PREFIX n2:	<http://bio2rdf.org/drugbank_resource:>
PREFIX n3:	<http://bio2rdf.org/drugbank_vocabulary:>
PREFIX n4:	<http://bio2rdf.org/drugbank:>
PREFIX n5:	<http://www.w3.org/2000/01/rdf-schema#> 
PREFIX rdf:	<http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX xsdh:	<http://www.w3.org/2001/XMLSchema#>

SELECT *
WHERE {
  ?s n3:x-kegg  ?o.
  ?s n5:label ?t
}  


This process yielded the following
results:
Total drugs queried: 10037
No DDI results: 5504
Total DDIs returned: 519193   (note to self: 'egrep "^dr:|INFO: Results: dr:" log.txt| wc -l' on log.txt)
Total DDIs skipped due to no DrugBank mapping: 478653
Original count of DDIs where both drugs could be mapped to DrugBank: 40540
Duplicates dropped (assumes no directionality to the PDDI): 19893
Final number of DDIs saved: 20647

So, that means that 20647 Kegg PDDIs are available for comparison. These are all in the file deduped-kegg-ddis.pickle. 



------------------------------------------------------------
PDDIs USED IN THE OSCAR EMR 
------------------------------------------------------------

* Dataset and Source Location

From the OSCAR documentation (https://sites.google.com/site/oscarusermanual/-oscar-emr/3-0-clinical-functions/3-7-1), we found the pointer that the Drug database for Oscar using is called Drugref2.


Drugref2 has its own application and sub-project under google code here:
https://code.google.com/p/drugref2/source/checkout

The SQL Dump of the dataset is available under the Oscar Project with date "2012-09-18"
http://sourceforge.net/p/oscarmcmaster/oscar/ci/master/tree/release/drugref.sql

SQL dump also available here; 
https://github.com/scoophealth/oscar/blob/master/release/drugref.sql
 

We downloaded the original interactions from Drugref2 to here; 
../original-drug-data/drugref2-DDIs-from-OSCAR.csv 


The following descriptions about Drugref are referenced from http://wiki.gnumed.de/bin/view/Gnumed/DrugRef

the Oscar implementation of drugref contains both Canadian DPD data and Crowther / Holbrook interaction data from the publication by these authors;
"Crowther, N. R., Holbrook, A. M., Kenwright, R., & Kenwright, M. (1997). Drug interactions among commonly used medications. Chart simplifies data from critical literature review. Canadian Family Physician, 43, 1972."

The original drugref involved a postgres database, python XML-RPC API, and php interface.

Oscar McMaster EMR reworked drugref into (what they refer to) as drugref2:
Oscar never did use the PHP interface (meant for maintaining a monolithic database)
Oscar originally imported Canadian DPD data into the original drugref schema, but found this slow 
subsequently let Canadian DPD data imports persist (as imported) in additional tables and added (a python?) DPD plugin to supply needed server-side data to the API (versioned in the oscar cvs hosted on sourceforge, but hosted for the most part on savannah)
added tables to speed searching
imported later-acquired Holbrook interactions data
ported to mysql
in 2009
 made a java version of the API (as an alternative to the pythonic API)
 implemented the database layer in JPA (Java Persistence API), which has made it easier to support both Postgresql and Mysql
 http://code.google.com/p/drugref2/source/checkout
 you can run the java drugref webapp on-top of an existing postgresql drugref database and it works
 

* mapping from ATC to RxNorm to DrugBank

Example query of ATC code to get RxNorm CUI using OMOP Standard Vocabulary 4.4: 

---
SELECT  C.concept_id    ingredient_concept_id,
        C.concept_name  ingredient_concept_name,
        C.concept_class ingredient_concept_class,
        C.concept_code  ingredient_concept_code
 FROM   concept          C,
        concept          D,
        concept_ancestor CA
 WHERE  d.concept_code = 'H03AA03'
   AND CA.ancestor_concept_id = d.concept_id
   AND  C.concept_id           = CA.descendant_concept_id
   AND  C.vocabulary_id        = 8
   AND  C.concept_level        = 2;
---


* Mapping from RxNorm to DrugBank using LinkedSPLs RDB April 2014:

---

SELECT * FROM `FDAPreferredSubstanceToRxNORM` INNER JOIN 
              `FDA_SUBSTANCE_TO_DRUGBANK_BIO2RDF` ON  
              `FDAPreferredSubstanceToRxNORM`.PreferredSubstance = `FDA_SUBSTANCE_TO_DRUGBANK_BIO2RDF`.PreferredSubstance 
WHERE 
`FDAPreferredSubstanceToRxNORM`.RxNORM IN ('http://purl.bioontology.org/ontology/RXNORM/6703','http://purl.bioontology.org/ontology/RXNORM/6691','http://purl.bioontology.org/ontology/RXNORM/14584','http://purl.bioontology.org/ontology/RXNORM/7514','http://purl.bioontology.org/ontology/RXNORM/7518','http://purl.bioontology.org/ontology/RXNORM/22656','http://purl.bioontology.org/ontology/RXNORM/6373','http://purl.bioontology.org/ontol ogy/RXNORM/24591','http://purl.bioontology.org/ontology/RXNORM/6529','http://purl.bioontology.org/ontology/RXNORM/7519','http://purl.bioontology.org/ontology/RXNORM/1005921')