Text mining to produce large chemistry datasets for community access
Download as PPTX, PDF2 likes635 views
The document discusses the development of large chemistry datasets through text mining for community access, highlighting various data models for predicting melting points and NMR data extraction. It covers the methodologies used, including potential quality controls for consistent data representation across databases like ChemSpider and PubChem. The work emphasizes collaboration across institutions for enhanced data accessibility and analytical capabilities in chemistry.
![Some modeling highlights
LibSVM grid search was used to select parameters in grid (ca
1.5 years of CPU-time optimization)
Largest model:
668k descriptors (MolPrint) ~ 0.2 trillions entries
Biggest model:
618Mb (Dragon descriptors)
Most accurate model: Consensus, average of 5 models
RMSE < 32°C for the drug like region, MP [50,250]°C](https://image.slidesharecdn.com/text-miningtoproducelargechemistrydatasetsforcommunityaccess-150818125540-lva1-app6891/85/Text-mining-to-produce-large-chemistry-datasets-for-community-access-16-320.jpg)
Text mining to produce large chemistry datasets for community access
- 1. Text-mining to produce large chemistry datasets for community access Valery Tkachenko1, Aileen Day1, Daniel Lowe2, Igor Tetko3, Carlos Coba4 , Antony Williams5 1 Royal Society of Chemistry, UK 2 NextMove Software, UK 3 HelmholtzZentrum München, Germany 4 Mestrelab Research, Santiago de Compostela, Spain 5 EPA, US ACS Fall 2015 Boston, MA August 17th 2015
- 3. ChemSpider
- 4. Refs - we live in linked world
- 5. Properties
- 7. Knowledge systems Datastore Raw data Data in process Data out process UI, API, Services, etc
- 8. RSC Archive – since 1841
- 10. Further work – properties and spectra mining
- 11. Text mining of the chemical documents Term Examples of text matched FromLiterature “lit.” MeltingPoint “mpt”, “melting point”, “m.p.” Qualifier “>”; “approximately” Value “75° C”, “200° F”, “one hundred degrees Celsius” Range “184-186° C”, “191.5 to 192.4° C” MeasurementE rror “50±° C” OutcomeQuali fier “decomp.”, “with decomposition”, “subl.” FromLiterature? MeltingPoint Qualifier? (Value | Range | MeasurementError) OutcomeQualifier?
- 12. Why MP? Used for water solubility prediction Yalkowsky equation: logS = 0.5 – 0.01(MP-25) – log Kow
- 13. Detecting suspicious melting points • Value was greater than 500° C • Value was a range wider than 50° C • Value was a range where the second temperature was lower than the first temperature
- 14. 300k Melting Point Datasets Bergström 277 Bradley 2886 OCHEM 22404 Enamine 21883 Patents 228079 data Bergström Bradley OCHEM Enamine Patents Tetko et al J. Chemoinformatics, in preparation
- 15. Melting point model: data distribution
- 16. Some modeling highlights LibSVM grid search was used to select parameters in grid (ca 1.5 years of CPU-time optimization) Largest model: 668k descriptors (MolPrint) ~ 0.2 trillions entries Biggest model: 618Mb (Dragon descriptors) Most accurate model: Consensus, average of 5 models RMSE < 32°C for the drug like region, MP [50,250]°C
- 17. Prediction error
- 18. NMR data • Extract from 1976-2014 USPTO applications *unknown – starts off with NMR: peak list (no nucleus) H 975543 C 56536 unknown 44306 F 9429 P 3241 B 91 Si 62 Sn 22 Se 11 N 8
- 19. NMR text mining • We can find and index text spectra:13C NMR (CDCl3, 100 MHz): δ = 14.12 (CH3), 30.11 (CH, benzylic methane), 30.77 (CH, benzylic methane), 66.12 (CH2), 68.49 (CH2), 117.72, 118.19, 120.29, 122.67, 123.37, 125.69, 125.84, 129.03, 130.00, 130.53 (ArCH), 99.42, 123.60, 134.69, 139.23, 147.21, 147.61, 149.41, 152.62, 154.88 (ArC)
- 20. NMR extracted by year of publication 0 500000 1000000 1500000 2000000 2500000 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 CumulativedistinctNMRextracted Year of Publication USPTO grants USPTO applications
- 21. NMR solvents 48.5% 38.3% 8.7% 1.1% 1.0% 1.0% 1.4% CDCl3 DMSO-d6 CD3OD D2O Acetone-d6 MeOD Others Others: CD2Cl2, CD3CN-d3, C6D6, Pyridine-d5, THF-d8, CD3Cl, dimethylformamide-d7, d1-trifluoroacetic acid, methanol-d3, acetic acid-d4, toluene-d8, sulfuric acid-d2, 1,1,2,2- tetrachloroethane-d2, CD3OCD3, dioxane-d8, 1,2-dichloroethane-d4
- 22. 1H-NMR frequency over time 0 Mhz 50 Mhz 100 Mhz 150 Mhz 200 Mhz 250 Mhz 300 Mhz 350 Mhz 400 Mhz 450 Mhz 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year of patent filing
- 24. 1H NMR (CDCl3, 400 MHz): δ = 2.57 (m, 4H, Me, C(5a)H), 4.24 (d, 1H, J = 4.8 Hz, C(11b)H), 4.35 (t, 1H, Jb = 10.8 Hz, C(6)H), 4.47 (m, 2H, C(5)H), 4.57 (dd, 1H, J = 2.8 Hz, C(6)H), 6.95 (d, 1H, J = 8.4 Hz, ArH), 7.18–7.94 (m, 11H, ArH)
- 25. 13C NMR (CDCl3, 100 MHz): δ = 14.12 (CH3), 30.11 (CH, benzylic methane), 30.77 (CH, benzylic methane), 66.12 (CH2), 68.49 (CH2), 117.72, 118.19, 120.29, 122.67, 123.37, 125.69, 125.84, 129.03, 130.00, 130.53 (ArCH), 99.42, 123.60, 134.69, 139.23, 147.21, 147.61, 149.41, 152.62, 154.88 (ArC)
- 26. Detecting suspicious NMR spectra • Last peak of NMR spectra is unannotated and: – All other peaks are annotated – Spectrum has 1 peak and is proton or unknown NMR
- 27. > <SuspiciousValue> true > <Value> 1H-NMR (400 MHz, d6-Acetone): 11.8-10.8 (brs, 1H), 7.78 Comments: Only the labile proton is reported in the spectrum. The other aromatic and aliphatic protons are completely missing in the spectrum.
- 28. > <SuspiciousValue> true > <Value> 1H-NMR (400 MHz, CDCl3): 6.85 (1H, d, J=7.8 Hz), 6.10 (1H, dd, J=7.8 and 2.2 Hz), 6.06 (1H, d, J=2.2 Hz), 4.66 (1H, m), 3.75 (4H, br s), 3.40 (2H, s), 1.97 Comments: There are only 11 protons reported in the spectrum whilst the molecule contains more than 50 protons.
- 29. Knowledge systems Datastore Raw data Data in process Data out process UI, API, Services, etc
- 30. Synthetic chemistry article Compounds Reaction Analytical Data Text and References
- 31. RSC Databases RSC Compounds RSC Reactions RSC Spectra RSC Crystals RSC Polymers RSC Materials RSC Assays RSC Algorithms RSC Models …and on…
- 32. Input pipeline Deposition Gateway Staging databases Compounds Reactions Spectra Crystals Materials Compounds Module Spectra Module Reactions Module Materials Module Textmining Module Module Web UI for unified depositions DropBox, Google Drive, SkyDrive, etc ELNs, templated data input Documents API, FTP, etc Raw data Validated data Staging databases All databases are sliced by data sources/ data collections and have simple security model where each data slice/ source is private, public or embargoed Etc Experiments Research
- 33. Output pipeline Compounds Reactions Spectra Crystals Documents Compounds API Reactions API Spectra API Crystals API Documents API Compounds Widgets Reactions Widgets Spectra Widgets Crystals Widgets Documents Widgets Data layer Data access layer User interface widgets layer Analytical Laboratory application User interface layer (examples) Electronic Laboratory Notebook Paid 3rd party integrations (various platforms – SharePoint, Google, etc) Chemical Inventory application ChemSpider 2.0
- 35. Compounds domain
- 36. Data quality issue and CVSP – Robochemistry – Proliferation of errors in public and private databases • ChemSpider • PubChem • DrugBank • KEGG • ChEBI/ChEMBL – Automated quality control system
- 37. Chemistry Validation and Standardization Platform
- 38. Reactions domain
- 39. Reactions domain
- 43. New Repository Architecture doi: 10.1007/s10822-014-9784-5
Editor's Notes
- #22: List of others probably isn’t completely comprehensive (solvent is free text!). 2 million spectra (from USPTO applications) have identified solvents
- #23: Excluded results < 1MHz and >1GHz (…mixing up Hz and MHz not uncommon!). Just to confuse things this is from the grant data while the previous data was from applications :-p
- #27: Extracted NMR spectrum is truncated as it finds the valid spectra up till before the error US20140378645A1 0057 typo? US20140378687A1 0195 missing open bracket
- #33: Change to add more database, rearrange
- #40: Information typically associated with reactions