New challenges in the management of chemical information. Service of cheminformatics

Transcription

1

2 New challenges in the management of chemical information Service of cheminformatics

3 Convert Intranet information to knowledge

4 bp LD50 pka Convert mp IC50 information to knowledge NMR IR logp ms Xray

5 physical property QSAR Convert bond length spectra information toxicity fragmentation to knowledge

6 Data

7 Some available databases ChemSpider ChemExper emolecules Zinc

8 Specialized databases CSD (Cambridge structural database) KNApSAcK AIST: Spectral Database NIST Chemistry web book

9 Downloadable databases Crystallography Open Database ChemBL DrugBank PDB Pubchem GDB-17 NMR shift DB KEGG Wikipedia Binding DB Mass bank OChem : Online chemical database

10 in 5 years? in 10 years? FTP USB File Server Print in 20 years? PDF Install proprietary software for each data format

11 Store correctly the chemical information Perennial and reprocessable! Only use ALIVE published DESCRIBED format no proprietary format no PDF, word, excel, chemical structures: molfile spectra: JCAMP-DX

12 Store correctly the chemical information bp : sign - low - high - pressure (20 torr) ; 10 (15 mmhg) ; >300 ; NMR : 1H NMR (CDCl3): δ = 0.84 (3 H, t, J = 7.4 Hz, CH3), 0.94 (3 H, t, J = 7.4 Hz, CH3), 1.23 [6 H, q, J = 6.9 Hz, P(O)OCH2CH3], 1.51 (4 H, m, CH2), 2.20 (1 H, sextet, J = 6.6 Hz, CH), 3.80 (3 H, s, OCH3), 4.01 [4 H, m, P(O)OCH2CH3], 4.63 (1 H, d, JH,P = 17.1 Hz, NCHP), 6.88 (2 H, d, J = 8.6 Hz, CH). Traceability, flexibility : nosql database

13 DB Image analysis Mass analysis Similarity search Predictions open data

14 Retrieving and processing data The need for a flexible tool

15 Pure HTML5 MacOS, windows, linux No installation Knowledge open-source (GitHub) No licence fee Teaching Research

16

17 How does it works?

18 Basic example : calculate a sum of 2 values onde code executor single value a b var c=a+b c

19 Demo

20 Chemical types - MF and SMILES onde code executor single value data var result = { type : mf, value : data } result C6H15N CCC(=O)C C6H15N O

21 Demo

22 Using external libraries onde code executor object editor mf var info=cc.analyzemf(mf); info ChemCalc

23 Demo

24 Molecular formula and mass

25 Isotopic distribution Isotope mass (Da) Abundance 35 Cl % 37 Cl % Cl + Cl2 + m/z m/z

26

27 ChemCalc Use of groups (Ph, Ala, Gly, Fmoc) HAlaGlyProOH Many components (separated by. ) C100. C110. C120 Specific isotope [ 13 C]100. [ 13 C]50[ 12 C]50 Modification of the isotopic abundance C{50,50}10C10 H 2 N O OH Specification of the charge HAla10OH +. HAla10OH ++. HAla10OH (H + )2 O Modification of a molecular formula HSer(H-1Ph)OH

28

29 Monoisotopic mass - most abundant isotopes m/z isotope % mass (Da) 1 H C N O MF mass (Da) ppm NO CH2O H2N CH4N C2H

30 Statistics of pubchem Search by EM and MF Pubchem SDF Visualizer FTP download Calculate MF MF stats Exact mass Webservice MongoDB

31

32 Contaminants / Custom database Two possible problems : Mass spectra may contain contaminants You want to look for a list of products A solution : Edit a list of molecular formula in a spreadsheet Match directly the list to your spectrum

33

34 Peptide fragmentation x3 y3 z3 x2 y2 z2 x1 y1 z1 R 1 H N O R 3 O H 2 N O R 2 N H H N O R 4 OH a1 b1 c1 a2 b2 c2 a3 b3 c3

35

36 Cheminformatics

37 From 2D drawing to conformations Draw Display 2D molfile conformers Server moloc 2D to 3D

38 Demo

39 Lipinski rule of 5 H bond donors 5 H bond acceptors 10 log P mw

40 Demo 1 Demo 2

41 Virtual combinatorial library The problem Search for products with specific properties : logp mw H donor H acceptor A solution R 2 R 3 R 4 R 1 Generate all possible molecules based on a template and fragments Predict the properties Filter using parallel coordinates N Vanderveen, J. R., Patiny, L., Chalifoux, C. B., Jessop, M. J., & Jessop, P. G. (2015). A virtual screening approach to identifying the greenest compound for a task: application to switchable-hydrophilicity solvents. Green Chemistry, 17(12),

42 Demo

43 Substructure search - Wikipedia

44 Substructure search? O O OH NH N O Cl NH 2 O Ertl, P., Patiny, L., Sander, T., Rufener, C., & Zasso, M. (2015). Wikipedia Chemical Structure Explorer: substructure and similarity searching of molecules from Wikipedia. Journal of cheminformatics, 7(1), 10.

45 c1ccccc1

46 Every night Find ChemBox DrugBox Extract SMILES Generate JSON Publish GitHub

47 Demo

48 Simple database of chemicals - open source malaria Store Google spreadsheet Display Live extraction (tsv) Prediction visualizer "Open source drug discovery: highly potent antimalarial compounds derived from the tres cantos arylpyrroles." ACS central science 2, no. 10 (2016):

49 Link to google docs To the visualizer

50 Validation dataset for ab initio project PubChem internal database Convert to XYZ, zip, download Filter molecules moloc add H 2D to 3D UI to browse and select

51 Demo

52 Image analysis

53 Image analysis : Bacteria in water Coliscan Easygel Classification Image analysis

54 Demo

55 Image analysis: IC50 [ ] Bioluminescent sensor proteins for point-of-care therapeutic drug monitoring Kai Johnsson et al., Nature chemical biology 2014, 10,

56 The problem : many steps [ ]

57 Demo

58 SEM / TEM image analysis TEM image Statistics Image analysis

59 Demo

60 Nuclear magnetic resonance

61 1 H NMR spectroscopy ppm

62 HOSE code 1.19 H 3 C 3.64 OH 2.87 O H 3 C O 1.24 CH 3 H 3 C 2.23 O OH 12.0 δ = = 1.81 O 3 H 3 C O CH 3 δ = = 2.13 δ = 2.03

63 NMR self learning algorithm - Ask Ernö Creating a NMR chemical shifts predictor without using chemical shifts!? Castillo AM, Bernal A, Dieden R, Patiny L, Wist J. Ask Ernö : a self-learning tool for assignment and prediction of nuclear magnetic resonance spectra. Journal of cheminformatics Dec 1;8(1):1-8.

64 1. Creating a dataset: molfile / spectrum 2341 molecules / NMR spectra

65 2. Number and kind of hydrogens d 3 c b a c b

66 3. Unambiguous number of protons

67 3. Peak picking and automatic integration Integration of the zones Removal of the NMR solvent / impurities

68 Demo

69 4. Assign non ambiguous protons and create corresponding HOSE codes

70 Ob 2 Iteration 8 Analysis 0 of the 2341 molecule / spectra set Predicted Chemical Shift 10 Observed Chemical Shift Iteration 9 HOSE 3 : 63 HOSE 4 : 382 HOSE 5 : Predicted Chemical Shift ach iteration of the training loop Test set of 298 molecules

71 Quality of prediction based on HOSE code level 10 HOSE 3 HOSE 4 10 Observed Chemical Shift Spheres Observed Chemical Shift Spheres Predicted Chemical Shift Predicted Chemical Shift 10 HOSE 5 Observed Chemical Shift Spheres Predicted Chemical Shift Fig. 6 Correlation between observed and predicted chemical shift values after learning for different sphere radius (iteration 9)

72 Prediction average errors NMRPredict Chemdraw 11 Spinus ACD/NMR Ask Erno 0% 25% 50% 75% 100% < 0.05 ppm < 0.15 ppm < 0.25 ppm < 0.35 ppm > 0.35 ppm

73 We need reliable structured data!

74 DB Image analysis Mass analysis Similarity search Predictions open data

75 Demo

76 Conclusions

77 Advanced visualization platform Solves scientific problems in the browser Pure HTML5 / javascript Easily reusable javascript library Research, service and teaching Over 150 tools in 3 years

78 Client / Server approach Server Client

79 Thanks! ChemCalc Alain Borel (Library EPFL) Laure Menin Michael Krompiec Marek Noga NMRdb Julien Wist Andrès Castillo Andrès Bernal script, visualizer, components Norman Pellet Michaël Zasso Daniel Kostro Jefferson Hernández Miguel Asencio Julien Wist Andrès Castillo

80