ChemSpider is an online database offering over 26 million unique molecules, utilizing open source components and crowd-sourced data curation. It addresses challenges in data quality and standardization, particularly with the InChI system, while focusing on accurate name-structure relationships for chemical entities. The project aims to standardize its entire database and enhance collaboration among data providers to improve the quality of chemical information available online.

1. ChemSpider – An Online Database and Registration System Linking the Web Antony Williams and Valery Tkachenko EBI Chemical Registry Systems Workshop, October 2011

2. www.chemspider.com

3. ChemSpider… >26 million unique molecules from >400 sources .NET, SQL Server and GGA Indigo toolkit Multiple Open Source components – Jmol, JSpecView, Balloon, OpenBabel, MediaWiki Slices of data are Open but the entire data collection is not Open Crowdsourced depositions and curations Uses InChIs for navigating and linking the web

4. Vancomycin

5. Vancomycin Search Molecular SKELETON Search Full Molecule

6. Full Skeleton Search: 104 Hits

7. Full Molecule Search: 4 Hits

8. ChemSpider… >26 million unique molecules from >400 sources .NET, SQL Server and GGA Indigo toolkit Multiple Open Source components – Jmol, JSpecView, Balloon, OpenBabel, MediaWiki Slides of data are Open but the entire data collection is not Open Crowdsourced depositions and curations Uses InChIs for navigating and linking the web Uses Names for navigating and linking the web

9. I want to know about “Vincristine” If all algorithms work then everything on the page is correct by default except the name-structure relationship!

10. Vincristine: Identifiers and Properties

11. Vincristine: Vendors and Sources Linked by Structure

12. Vincristine: Patents Linked by Name

13. Vincristine: Articles Linked by Name

14. ORIGINAL ChemSpider “ Create a system for linking and navigating databases on the web ” Use the power of InChI , and the proliferation of InChIs in databases, to make connections Developed on .NET and SQL Server for speed of implementation and existing skill sets Seeded with PubChem database of 10.5M chemicals and expanded using other sources to 20M

15. How do we build it? We deal in Molfiles or SDF files – with coordinates Deposit anything that has an InChI – we support what InChI can handle, good and bad Standardization based on “InChI standardization” InChIs aggregate (certain) tautomers We link out to external sites using their IDs

16. InChIs – both on ChemSpider

17. Downsides of InChI InChI was a moving target (multi versions) but overall worked as planned. Good for small molecules – but no polymers, issues with inorganics, organometallics, imperfect stereochemistry. ChemSpider is “small molecules” InChI used as the “deduplicator” – FIRST version of a compound into the database becomes THE structure to deduplicate against…

18. Side Effects of InChI Usage

19. SMILES by comparison…

20. Side Effects of InChI Usage

21. Standardization Issues Depiction based on molfile

22. Downsides of Overall Approach Meshing data together based on InChIs worked for simple molecules 2D layout errors inherited or limited by algorithm Complex molecules that are meant to be the same thing were NOT deduplicated. Compounds differing by one stereocenter, named the same, meant to be the same, are not the same

23. Yohimbine

24. Originally 15 compounds “called” Yohimbine 54 Skeletons for Yohimbine

25. ChemSpider as an Aggregator ChemSpider has inherited many errors, and it continues but we are way more careful now with pre-filtering Cannot deposit chemicals without an InChI Deprecated compounds remain deprecated Curated name-structure relationships do NOT remove the related structure If Taxol is removed from 20 asserted “incorrect structures” those compounds remain in the database

26. Chemistry Databases on the Internet Some public databases are “trusted” as primary sources Trust is granted without investigation or understanding of the content What do we know about some of the online resources?

27. PHYSPROP Database The freely downloadable database under the EPI Suite prediction software Very Basic filters suggest data quality issues

28. The Stereochemistry challenge. 12500 chemicals with “missed” stereo

29. Searches on ChemSpider Most searches are text-based: people searching for information about known chemicals Creating accurate name-structure dictionaries is critical

30. NIST Webbook

31. PubChem

32. NPC Browser http://tripod.nih.gov/npc/

33. NPC Browser http://tripod.nih.gov/npc/

35. NPC Browser http://tripod.nih.gov/npc/

36. Synonyms on PubChem 1,3-DICHLORO-PROPAN-2-ONE (2R,3R)-Butanediol bis(methanesulfonate) Ethyl-1-propenyl ether, mixture of cis and trans PSS-[2-[(Chloromethyl)phenyl]ethyl]-Heptaisobutyl substituted 1-Chlorobenzylethyl-3,5,7,9,11,13,15-heptaisobutylpentacyclo [9.5.1.1(3,9).1(5,15).1(7,13)]octasiloxane

37. Synonyms on PubChem

38. Data Proliferation

44. What is meant by a name?

45. Choose a Starting Point

46. “ The First 10”

47. What is getting into Our Databases? Large aggregators are inheriting junk data Data HAS proliferated from ChemSpider through PubChem – in process of deprecating and redepositing A lot of data is for chemicals that will never exist (probably)

48. Standardization of Patent Data???

49. Standardization of Patent Data???

50. WYSIWYG compounds

51. WYSIWYG compounds

52. Text Mining Chemical Name Errors

53. “ DPA”

54. All aggegators suffer dilution!

55. Structures have timelines

56. Name-Structure Dictionaries…

57. Depiction for Humans

58. Human Depiction versus Algorithms

59. Human Depiction versus Algorithms

60. Identifier Dictionaries Reciprocal curation processes…share curation with each other. If a database has a compound already then use InChiKeys to match “suggested” validation against the compound. A series of “added” and “removed” synonyms against InChIKeys for matching.

61. Proof of Concept Data Curation Sharing

62. Structure Validation using feed Look for approved synonyms Compare feed InChIKey with database InChIKey If different, flag for inspection

63. Open PHACTS : partnership between European Community and EFPIA Freely accessible for knowledge discovery and verification. Data on small molecules Pharmacological profiles Pharmacokinetics ADMET data Biological targets and pathways Proprietary and public data sources.

64. Adopting Modified FDA Rules As already used by ChEMBL…

65. Nitro groups

66. Salt and Ionic Bonds

67. Ammonium salts

68. Parent and Child Chemical entities reduced to primary component plus relationships salt forms solvates combinations

69. ChemSpider Standardization Entire ChemSpider database will be standardized using modified FDA rule set Original Molfiles will be standardized and all properties (predicted properties, SMILES, InChIs, Names) will all be regenerated Standardization procedures automatically applied to all future depositions

70. Project Status Standardization pipelining process initiated Rule implementation and checking – iterative work with Open PHACTS pharma members Data model development to support parent-child relationships In dialog with the FDA about latest form of recommendations

71. Conclusions ChemSpider has an important role in quality data Crowdsourced deposition, validation and curation works but low engagement to date Standardization of our entire backfile is necessary Designing the standardization processes with input from pharma and general chemists is necessary

72. Acknowledgments The ChemSpider team Our data providers, depositors, collaborators and curators Software providers – OpenEye, ChemDoodle, ACD/Labs, GGA Software, Open Source (Jmol, JSpecView, OpenBabel)

73. Thank you Email: williamsa@rsc.org Twitter: ChemConnector Blog: www.chemspider.com/blog Personal Blog: www.chemconnector.com SLIDES: www.slideshare.net/AntonyWilliams