1. Purpose of the visit
The purpose of this 5 day visit was to develop a model to verify the organic identity of feeds used for laying hens by analyzing feeds by Near Infrared Spectroscopy (NIR). The model will be compared with a model already developed at the home institution by using fatty acid fingerprinting.
Furthermore, this visit allowed the fellow to get familiarized with the NIR technique and the statistical treatment of NIR data.
2. Description of the work carried out during the visit
First, an introduction to the NIR analysis was given by Prof. Garrido-Varo to the fellow. NIR analysis of feeds was conducted during the 2nd and 3rd day. Samples used in this stay were collected by the RIKILT, in the Netherlands, and were sent to Cordoba for the NIR measurement. Samples consisted of 99 different feeds (set1, 50 samples: set2, 49 samples) used for laying hens. They included: 36 organic feeds, and 63 conventional feeds (24 used for free range hens, 24 used for barn hens, 13 used for caged hens, and 2 used for barn hens laying high omega-3 eggs).
These feeds had been grinded to 0.5 mm particle size. NIR measurements were performed using a FOSS NIR system 6500 SY-I, equipped with a spinning module, working in reflectance mode, in the spectral range 400-2500 nm, taking readings every 2 nm (FOSS NIRSystems, Inc., Laurel, MD, USA). Measurements were taken using standard ring cups (diameter of 3.75 cm). After NIR analysis, the statistical treatment of the data was started, and it is being continued at the home institution by using Pirouette 4.0 (Infometrix, Seattle, USA) software. Firstly, several data pretreatment was assessed: SNV, 1st and 2nd derivatives and smoothing. Principal Component Analysis (PCA) was performed with the most successful pretreatments to investigate if there was any natural clustering within the data, and to detect possible outlier samples. As the identity of feeds was known (according to the production system they were used for), a supervised clustering technique, Partial Least Squares-Discriminant Analysis (PLS-DA), was applied. Models created were validated by cross validation: 10% of the samples was removed each time, the model was built again with the remaining samples, and it was used to predict the identity of the removed samples. This was repeated until predictions were obtained for all samples.
During the stay in Cordoba, the fellow got familiarized with the NIR analysis, NIR instruments and the data pretreatments performed with NIR spectra. The stay consisted only on 5 working days, so later, the statistical analysis is going oncurrently at the home institution.
3. Description of the main results obtained
Multivariate statistical techniques were applied to the NIR spectral data. Samples were explored by PCA, with the most promising data pretreatments (SNV, 1st der and smooth). PCA model explained more than 98% of variance with only 2 factors. PCA (none preprocessing) showed a tendency of organic and conventional classes to be separated. This made that the application of a supervised classification techniques was promising. PLS-DA was applied to samples belonging to set 1, and the model was first cross validated. The PLS-DA after a SNV, 1st derivative (5 points) and smoothing (5 points) showed quite successful results when it was cross validated (leaving 5 out). To externally validate the model, the model was applied to samples from set 2. The number of correct classifications was quite successful, although some false positive appeared (Table 2). These results can be improved with an optimization of data pretreatment and PLS-DA parameters, which is currently being performed.
Results emerging from the final model will be compared with results from classification models developed at RIKILT for the verification of organic feeds using fatty acid fingerprinting.
The STSM has been very successful because it has allowed to obtain the NIR spectral data of feeds belonging to an ongoing project at RIKILT, and the fellow got familiarized with the NIR analysis. This STSM only lasted 5 days, thus the statistical treatment of the data is currently performed at the home institution, and the model presented here is being optimized, both by modifying the parameters during pretreatment of the data, as well as by optimizing the PLS-DA parameters.
. • Future collaboration with host institution
After the STSM, the collaboration with the host institution will be continued because the statistical analysis is currently being performed. Also, this STSM opened the possibilities for collaborations between both institutions in future projects.
• Projected publications/articles resulting or to result from the STSM
When the final model is developed, results will be presented in form of a poster in an international conference. Also, they will be included in a publication in a peer reviewed journal.
• Confirmation by the host institute of the successful execution of the mission Dr. Alba Tres moved for 5 days to the None-destructive Sensors Unit form the University of Córdoba, to analyze feeds by NIR. Analysis were performed, and the statistical treatment of the data was started. Pretreatment of the NIR data started at the host institution. Model development and the optimization of the parameters is going on at the home institution. The preliminary results are promising, so when optimized they will be publishable in a peer-reviewed scientific journal. This STSM has been a fruitful collaboration, which will be maintained in the future.