Prédiction de la composition en acide gras des amandes des noix de palmier à huile par spectroscopie proche infra rouge (SPIR)

•Step 1: Cross validation preprocessing and outliers ➢103/20 samples for calibration/validation set ➢Spectra data were regressed with each FA content by PLS-R (Partial Least Square R regression) Models performance evaluated in cross validation on RMSECV (Root Mean Square Error of Cross Validation) to choose the best: - preprocessing spectra - NIR region - Latent Variable (LV) •Step 2: Model validation by test set For each FA, the best model was used to predict the test set with the selected preprocessing spectra and the corresponding optimal LV number Accuracy of the FA content prediction depends on the spectra range Predicting the fatty acid (FA) composition of kernels using near-infrared (NIR) spectroscopy V. Pomiès 1,2,V. Vaissayre 3,M. Delodde, H. Domonhedo 4, F. Jacob 5,S. Tisné 1,2,S. Dussert 3,G. Chaix1,2,6 1 CIRAD, UMR AGAP Institut, F-34398 Montpellier, France, 2 UMR AGAP Institut, Univ Montpellier, CIRAD, INRAE, Institut Agro, Montpellier, France, 3 DIADE, Univ Montpellier, IRD, CIRAD, Montpellier, France, 4INRAB, CRAPP, Pobè, Benin,5 PalmElit SAS, 34980 Montferrier sur Lez, France,6 ChemHouse, Research Group, Montpellier, France Contact: virginie.pomie[email protected] wwwcirad.fr 1 Plant materials and Methods 2 NIR Spectra: acquisition - validation 4 NIR calibration models 3 Chemical analysis 5 Conclusion and applications •2 populations, La Mé and Yangambi, each represented by 11 and 12 crosses chosen to cover intra-population diversity of both populations (CRAPP, Pobè, Benin) •254 palm trees, 8-10 fruits/tree, fruit pulp removed and stored in dry place •Sample preparation: Palm nut cut in half with a bandsaw •Material: ASD Labspec (400-2500 nm) with optical fiber •2 spectra acquisitions / half kernel Total: 4780 NIR spectra •Average of the 2 spectra •Principal Composant Analysis (PCA) on 1100 to 2350 nm: Variability / intra population •Heritability of spectral measurement (h² ≃0.6) - Fruits representative of the tree (Vi/VP) - Trees representative of the progeny (Vprog/Vi) Wavelength (nm) Tree variance/total variance (Vi/Vp) in green, progeny variance/individual variance(Vprog/Vi) in red •123 randomly selected palms representative of both intraand inter-population diversities •Pool half kernels/tree •2 lipid extractions per sample •FA methyl esters analyzed using Gas Chromatography (Agilent) •2 groups: major FA: C12, C14, C16 and C18:1-9 (32-50%) minor FA: C6, C8, C10, C18 and C18:2-6 (≃3%) •Variability of fatty acid composition Yangambi: high content C14 and C16 La Mé: high content C8, C10 and C12 PCA chemicals value /sample and genetic population, La Mé in green and Yangambi in red Best prediction model by test set, N=20 •High variability in fatty acid composition depending on genetic background •Good FA prediction by Nir tool •High throughput and cost-effective phenotyping method •Breeders use: Marker Assisted Selection for genetic variation •Genotype selection for kernel oil use in cosmetic applications Population Heritability estimation LV: Latent variable; RMSEP: Root Mean Square Error of Prediction; RPD: Ratio Performance Deviation (SD/RMSEP) S S LC LC C8 SC LC