1. I NDEPENDENT C OMPONENTS A NALYSIS I NDEPENDENT C OMPONENTS A NALYSIS Applications of ICA Douglas N. Rutledge, Delphine Jouan-Rimbaud Bouveresse douglas.rutledge@agroparistech.frdouglas.rutledge@agroparistech.fr delphine.bouveresse@agroparistech.frdelphine.bouveresse@agroparistech.fr

2. Information Hidden in MIR Spectra Source : R.Aries, D. Lidiard, R. Spragg, Spectrosc. Internat., 2(3) 41-44

3. ICA on 100 Mid-Infrared spectra Samples : a single polystyrene film Mid-IR spectra taken at end of production line Acquired from 4000cm -1 to 400cm -1, at 1 cm -1 Pre-treatment of spectra : - normalised between 0 and 1 - neither centred nor standardised

4. PC2  CO 2 & H 2 O Results of a PCA on the polystyrene MIR data

5. PC6 Interferogram & derivative PC5 Interferogram & derivative

6. PC5-PC6 Scores Plot

7. IC2/7 signal looks like water vapour spectrum Due to variations in moisture content of air

8. IC3/7 signal looks like spectrum of CO 2 Due to variations in CO 2 content of air

9. IC4/7 signal looks like beats of a simple interferogram Due to variations in optical path of polystyrene sample ?

10. IC7/7 signal looks like first derivative of average MIR spectrum Due to one spectrometer (N°61) being badly adjusted (frequency shift)

11. Elimination of artifacts Monitoring changes in oils during heating using 3-D Fluorescence Spectroscopy

12. Rayleigh scattering Raman scattering 3-D Fluorescence Spectra of Oils during Heating

13. IC1 = Rayleigh + Raman

14. IC2 = Polyphenols

15. IC3 = Chlorophyll…

16. Antioxydant influence of catechin on rats after hyperlipidic diets, monitored using a LC-MS based metabolomic approach

17. The data Samples Male Wistar rats (n = 8/group) fed for 6 weeks normo- (5% diet w/w) or hyper- lipidic (15 and 25%) diets (CT05 / HF15 / HF25) With or without catechin supplementation (0.2% w/w) (NP / PP) (polyphenolic antioxidant - helps prevent inflammatory and coronary diseases) Urines collected 17 and 38 days after diets were given (T17 / T38) Technique Analysed by mass spectrometry on a LC-QToF (m/z 100-1000; positive ionization) (HPLC Alliance 2695, Symmetry® RP18 column, Micromass Qtof-Micro / Waters) Pretreatment Intensities of peaks transformed to log(X+1) Variables sorted in order of decreasing variance (later as function of Retention Time)

18. Log(raw LC-MS data)

20. Independent Components Analysis Extract “pure signals” from observed mixtures - “pure signals” => “Loadings” - “proportions” of “pure signals” to mixtures = “Scores”

21. ICA on Log(Data)

27. IC2 “Loadings”

28. IC3 “Loadings”

29. IC5 “Loadings”

30. Mid-Infrared analysis of edible oils heated at 190° for 3 hours

31. ICA applied to Mid-Infrared spectra 180 spectra acquired every 3 minutes over 3 hours during flat heating at 190°C

32. PCA Loadings

33. PCA Scores

34. ICA Signals

35. ICA Scores

36. Mid-IR spectrum of CO 2

37. Scores of samples on IC9 as a function of heating time

38. Mid-IR spectrum cis-trans isomerisation

39. Scores of samples on IC2 as a function of heating time

40. ICA applied to Raman hyperspectral images Hyperspectral images acquired for an authentic and a suspect pharmaceutical pill M Boiret, D N Rutledge, N Gorretta, Y-M Ginot, J-M Roger Utilisation de la microscopie Raman et des methodes chimiometriques pour la detection de comprimes contrefaits, SpectrAnalyse, 2014, in press

41. Images collected using a PerkinElmer RS400 system Microscope coupled to spectrophotometer with 785nm 400mW laser CCD sensor (Charge-Coupled Device) Sample on a motorized stage with a pitch of 50 microns Raman spectra acquired from 3200cm -1 to 100cm -1 Spectral resolution 2 cm -1 26 000 spectra on a surface of about 8mm * 8mm

42. ICA_by_Blocks applied to authentic and suspect pills

43. ICA Proportions ICA Signals & Reference spectra R=0.99 R=0.98 Authentic pharmaceutical pill

44. ICA Proportions ICA Signals Suspect pharmaceutical pill

45. ICA Signals & Reference spectrum (Metformine) Suspect pharmaceutical pill Compare ICA signals with spectral database ICA Signals & Reference spectrum (Avicel) R=0.96 R=0.99

46. ICA can be applied to data usually analysed using PCA Contributions of variables are easier to interpret CONCLUSION