transformative utility of inchikey searching in the mother of all databases
DESCRIPTION
From BioIT Workshop "A Bar Code for Chemical Structures: Using the InChI to Transform Connectivity between Chemistry, Biology, Biomedicine and Drug Discovery" http://figshare.com/articles/BioIT_Workshop_2014_Chem_Bio_via_InChI/1063314 Update June. Workshop attendees had access to all the slide sets via CHI. Some are on slideshare (e.g. from Antony Williams) but I have merged the sets into a PDF in the figshare link above. Abstract: Google indexing of the InChIKey (IK) has turned the web into a de facto chemical database with well over 50 million unique entries (PMID:23399051). The first block of the IK encodes molecular skeleton that can be used to give maximum recall of related structures. For example, Google searching XUKUURHRXDUEBC from atorvastatin displays ~200 low-redundancy links in ~0.3 sec with a low false-positive rate . These include most major databases and less familiar but valuable sources. The simplicity of the IK makes it useful for those less familiar with chemical searching. Advanced Google Search can be used to filter results, image searching gives complementary coverage and there are also hits in Google Scholar. IK searching thus becomes powerfully enabling for reciprocal document-to-database joins from legacy text tombs including over 50 years of biology < > chemistry. Open tools such as chemicalize.org can generate of IKs from patents, papers, abstracts or web pages. Open Drug Discovery data on tested, synthesized or even proposed compounds, can be globally connected in real-time by surfacing IKs in open laboratory notebooks, Wikis, blogs, Twitter, figshare etc. Following the ChemSpider precedent the IUPHAR/GTP database offers users IK Google searches from all ligand entries including peptides.TRANSCRIPT
1
www.guidetopharmacology.org
The transformative utility of InChIKey searching in the Mother of all Databases
(a.k.a. Google)Chris Southan
IUPHAR/BPS Guide to PHARMACOLOGY Web portal Group, Centre for Integrative Physiology, School of Biomedical Sciences, University of Edinburgh,
Hugh Robson Building, Edinburgh, EH8 9XD, UK. [email protected]
2
Outline
• Introduction: the atorvastatin example• Chem-to-bio context• IK stats and estimates • Extracting IKs from documents • IK database-to-database• Open Source malaria drug discovery as a testbed• Caveats and future prospects
3
The precedent
InChI as a web index for molecules
“We have now discovered, serendipitously, that these InChIs have been comprehensively and accurately indexed by the Google search engine. From preliminary exploration it appears that every known document in which an InChI appears has been indexed and that all are retrievable by standard queries with virtually 100% precision. This means that standard Web-based indexers, without any alteration, are capable of acting as completely precise chemical search engines. Although we have many years of developing chemistry on the web, this was an unexpected and very welcome finding”
Murray-Rust et al. 2004 http://lists.w3.org/Archives/Public/public-swls-ws/2004Oct/att-0019/
4
IK example: atorvastatin and metabolites
5
Fast and clean results
parentpara-hydroxy
ortho-hydroxy
6
Inner layer XUKUURHRXDUEBC image search
7
Making the chem < > bio join
BiochemistryMedicinal chemistry
ToxicologyChemical biology
Systems pharmacologyMetabolomicsDrug discoveryPharmacology
Chemogenomics
InChIKey
8
Getting biology out of text-tombs is not easy;Getting chemistry out is even more difficult
9
Why chem < > bio joining is difficult
• The majority of chemistry embedded in biological reports is specified as semantic names or images
• The MeSH to PubChem connectivity is patchy• Biologists use sequence database accession numbers, ontologies
and gene names widely but chemists rarely use open chemical database IDs
• Most bioactive chemistry in text does not have direct connectivity to databases (unlike GenBank/RefSeq/UniProt < > PubMed)
• Nat.Chem.Biol. is the only bio-journal that mandates PubChem reciprocal linking
• Most authors don’t engage with surfacing and connectivity (e.g. becoming PubChem submitters and/or figshare data depositors)
• Chemists and biologists tend not to communicate easily• GenBank started in 1982, PubChem in 2004• Inventors/authors under-cite their own medicinal chemistry patents
10
So how many IKs has Google indexed ?
• PubChem ~ 50 million • ChemSpider ~ 30 million • PubChem from patents (all sources) ~ 15
million• PubChem journal sources (PubMed + ChEMBL)
~ 1 million• Web sources outside the above (no idea) • Open ELNs (no idea)
Guestimate 60 million-ish
11
Databases < > documents:IK Googling facilitates reciprocal linking
Abstracts
Patents
Papers
15 mill
0.2 mill (mainly MeSH)
0.9 mill (ChEMBL)
12K
12
IKs with data-supported bioactivity (>biology)
• GVKBIO Online Structure Activity Relationship Database (GOSTAR ) = 6.3 million with SAR data from patents and literature (not tagged in PubChem)
• Thomson Pharma = 4.2 million selected examples from patents and literature
• PubChem BioAssay “active” = 0.93 million • ChEMBL (in PubChem) = 0.88 million • Thomson Pharma (2013 only) = 0.27 million• PubMed = 0.23 million • MeSH “pharmacology” = 12,719• INN or USAN = 10,707• Union of last two above = 19,334 intersect = 4,092• Prous (Thomson) Drugs of the Future = 7,218• DrugBank approved (via SIDs) = 1,504
Guestimate for chemistry with a useful level of solubility, stability, specificity and potency (e.g. < 250 nM) in biological systems ~ 0.5 million IKs (but of course we also need low potency and inactives for controls and SAR)
13
IKs and the representational hextet used in documents and databases
14
Extracting IKs from documents: OPSIN
15
Extracting IKs from documents: chemicalize.org
16
Extracting IKs from documents OSRA
17
Extracting IKs from documents: sketchers
18
IK call-outs in dbs: extending the link reach
19
Modified peptides/big stuff: connection where similarity struggles
http://www.guidetopharmacology.org/GRAC/LigandDisplayForward?ligandId=2532
20
OSM drug discovery: test bed for open data surfacing and connecting chem > bio
• Team are exploring chemistry surfacing/sharing in real time (e.g. ELNs, Wiki, Github, ChEMBLMalaria for project updates)
• Converted to IK utility (after the necessary evangelizing) • Global antimalarial drug R&D (open and closed) exemplifies
full range of connectivity issues that IK surfacing can potentially ameliorate
21
Actively unlocking IK connections
22
Name > structure > biology: missing links
23
Where the IK connects……
24
Chemicalize.org: 413 strucs/IKs from WO2011086531
CID 53311393 ->
25
WO2011086531 >chemicalize.org > SAR IC50s > figshare
surfaces and connects (e.g. PubChem)
26
Share structures via open MyNCBI
http://www.ncbi.nlm.nih.gov/sites/myncbi/collections/public/1zWhcobieZbIouGfUdsdbHek5/.
27
DIY surfacing of name < > IK connections
28
Caveats and risks for IK Googling
• Ranking heuristics are opaque and change• Results shift on short time scales (i.e. irreproducible)• No API (or good search result set parsers) • Don’t ignore corroborative searches in well-structured
databases• Searching common IKs is not generally useful (but can filter)• No good for similarity searching on their own (but you can
intersperse with similarity approaches)• In the relentless war between good and evil (Google verses
the SEO Dark Side) dodgy chemical suppliers are always pushing
• There may be future risks of common chemistry swamping• Names, SMILES or even IUPAC strings may sometimes give
Google hits where the IK misses (because its not there)
29
What does the future hold /need ?
• For manual searching Googling the IK is the “first stop shop”• InChI world-domination is proceeding• Inexorable increase in full-text, open access journals and crawled
open repositories (e.g. figshare)• Journals must encourage author chemistry mark-up to include the IK• More biologists getting into chemistry connections and databases• Boutique bioactive chemistry databases becoming more discoverable• SureChEMBL will improve image handling and get crawled• RSC Journal Archive > ChemSpider• ContentMine (Murry-Rust et. al.) 100 million facts, including journal-
extracted chemical structure streaming• More Open (Source) Drug Discovery > Google crawled ELNs with IKs• Wider community use of Chemicalize.org for targeted extractions• New IK via source expansion in ChemSpider and PubChem
30
Thanks and Questions
31
Extras
32
Abstract
Abstract: Google indexing of the InChIKey (IK) has turned the web into a de facto chemical database with well over 50 million unique entries (PMID:23399051). The first block of the IK encodes molecular skeleton that can be used to give maximum recall of related structures. For example, Google searching XUKUURHRXDUEBC from atorvastatin displays ~200 low-redundancy links in ~0.3 sec with a low false-positive rate . These include most major databases and less familiar but valuable sources. The simplicity of the IK makes it useful for those less familiar with chemical searching. Advanced Google Search can be used to filter results, image searching gives complementary coverage and there are also hits in Google Scholar. IK searching thus becomes powerfully enabling for reciprocal document-to-database joins from legacy text tombs including over 50 years of biology < > chemistry. Open tools such as chemicalize.org can generate of IKs from patents, papers, abstracts or web pages. Open Drug Discovery data on tested, synthesized or even proposed compounds, can be globally connected in real-time by surfacing IKs in open laboratory notebooks, Wikis, blogs, Twitter, figshare etc. Following the ChemSpider precedent the IUPHAR/GTP database offers users IK Google searches from all ligand entries including peptides.
33
Patent SAR from WO2011086531:Collating activities via SureChemOpen
CID 53311393 >
34
Triaging document or webpage chemistry
• Identify the structure specification types– Semantic names (all sources)– Code names (press releases, papers and abstracts) – IUPAC names (papers, patents and abstracts)– Images (papers, patents, & Google images)– SMILES (open lab books)– InChi strings (open lab books)– SDF files (open lab books, & github)
Convert these to a structure (e.g. SDF, SMILES, InChI) then:– Search InChIKey in Google– Search major databases– Compare extracted sets for intersects and diffs – Extend exact match connectivity with similarity
searching
35
Orthogonal joining
36
Triage example: a new antimalaria
The MMV390048 code name is linked to an image in press reports but is PubChem and PubMed -ve