TICTAC - Target illumination clinical trials analytics with cheminformatics

Mining ClinicalTrials.gov via AACT-CTTI-db for target hypotheses, with strong cheminformatics and medical terms text mining, powered by NextMove LeadMine and JensenLab Tagger.

Dependencies

• ClinicalTrials.gov
• AACT-CTTI-db
• NextMove LeadMine
• JensenLab Tagger.
• BioClients
• IDG Pharos/TCRD
• PubChem REST API
• ChEMBL REST API
• ChEMBL webresource client (Python client library).
• nextmove-tools

About AACT:

• AACT-CTTI database from Duke.
  • CTTI = Clinical Trials Transformation Initiative
  • AACT = Aggregate Analysis of ClinicalTrials.gov
• According to website (accessed June 2022), data is refreshed daily.
• AACT structure changed in November 2021, reflecting newer ClinicalTrials.gov API.
• Identify drugs by intervention ID, since may be multiple drugs per trial (NCT_ID).

References:

• AACT Data Dictionary, which references https://prsinfo.clinicaltrials.gov/definitions.html and https://prsinfo.clinicaltrials.gov/results_definitions.html.
• The Database for Aggregate Analysis of ClinicalTrials.gov (AACT) and Subsequent Regrouping by Clinical Specialty, Tasneem et al., https://doi.org/10.1371/journal.pone.0033677 (2012).
• The Clinical Trials Transformation Initiative. One Decade of Impact. One Vision Ahead, Clinical Trials (2018).
• How to avoid common problems when using ClinicalTrials.gov in research: 10 issues to consider, Tse et al., BMJ 2018; 361, https://doi.org/10.1136/bmj.k1452 (2018).
• See also: https://www.ctti-clinicaltrials.org/briefing-room/publications

Text mining, aka Named Entity Recognition (NER)

• Chemical NER by NextMove LeadMine.
• Disease NER by JensenLab Tagger.

Purpose:

• Associate drugs with diseases/phenotypes.
• Associate drugs with protein targets.
• Associate protein targets with diseases/phenotypes (via drugs).
• Predict and score disease-target associations.

Drugs may be experimental candidates.

AACT tables of interest:

Table	Notes
studies	titles
keywords	Reported; multiple vocabularies.
brief_summaries	(max 5000 chars)
detailed_descriptions	(max 32000 chars)
conditions	diseases/phenotypes
browse_conditions	MeSH links
interventions	Our focus is drugs only among several types.
browse_interventions	MeSH links
intervention_other_names	synonyms
study_references	PubMed links
reported_events	including adverse events

Overall workflow:

See top level script Go_tictac_Workflow.sh.

1. Data:
2. Go_aact_GetData.sh - Fetch data from AACT db.
3. Go_pubmed_GetData.sh - Fetch PubMed data from IDG TCRD.
4. Go_jensenlab_GetData.sh - Fetch dictionary data from JensenLab.
5. LeadMine:
6. Go_aact_NER_leadmine_chem.sh - LeadMine NER, CT descriptions.
7. Go_pubmed_NER_leadmine_chem.sh - LeadMine NER, referenced PMIDs.
8. Tagger:
9. Go_aact_NER_tagger_disease.sh - Tagger NER, CT descriptions.
10. Cross-references:
11. Go_pubchem_GetXrefs.sh - PubChem IDs via APIs.
12. Go_chembl_GetXrefs.sh - ChEMBL IDs via APIs.
13. Results, analysis:
14. tictac.Rmd - Results described and analyzed.

Association semantics:

• keywords, conditions, studies and summaries: reported terms and free text which may be text mined for intended associations.
• descriptions: may be text mined for both the intended and other conditions, symptoms and phenotypic traits, which may be non-obvious from the study design.
• study_references: via PubMed, text mining of titles, abstracts can associate disease/phenotypes, protein targets, chemical entities and more. The "results_reference" type may include findings not anticipated in the design/protocol.
• interventions include drug names which can be recognized and mapped to standard IDs, a task for which NextMove LeadMine is particularly suited.
• LeadMine chemical NER also resolves entities to structures via SMILES, enabling downstream cheminformatics such as aggregation by chemical substructure and similarity.

NextMove

Running NextMove Leadmine NER via nextmove-tools.

$ java -jar ${LIBDIR}/unm_biocomp_nextmove-0.0.1-SNAPSHOT-jar-with-dependencies.jar
usage: LeadMine_Utils [-config <CFILE>] [-h] -i <IFILE> [-idcol <IDCOL>]
       [-lbd <LBD>] [-max_corr_dist <MAX_CORR_DIST>] [-min_corr_entity_len
       <MIN_CE_LEN>] [-min_entity_len <MIN_E_LEN>] [-o <OFILE>]
       [-spellcorrect] [-textcol <TEXTCOL>] [-unquote] [-v]
LeadMine_Utils: NextMove LeadMine chemical entity recognition
 -config <CFILE>                     Input configuration file
 -h,--help                           Show this help.
 -i <IFILE>                          Input file
 -idcol <IDCOL>                      # of ID input column
 -lbd <LBD>                          LeadMine look-behind depth
 -max_corr_dist <MAX_CORR_DIST>      LeadMine Max correction (Levenshtein)
                                     distance
 -min_corr_entity_len <MIN_CE_LEN>   LeadMine Min corrected entity length
 -min_entity_len <MIN_E_LEN>         LeadMine Min entity length
 -o <OFILE>                          Output file
 -spellcorrect                       LeadMine spelling correction
 -textcol <TEXTCOL>                  # of text/document input column
 -unquote                            unquote quoted column
 -v,--verbose                        Verbose.