Agricultural University of Athens
Διερεύνηση της εφαρμογής της φασματοσκοπίας ανάκλασης εγγύς υπερύθρου στην πρόβλεψη της διαιτητικής αξίας σύνθετων ζωοτροφών σε κόνικλους
Abstract
dc:descriptionThe determination of digestibility has a key role in the nutritive evaluation of animal feed, but due to the cost and time of the classical determination methods it cannot be performed in real time, where it is of particular importance. Near-infrared reflectance spectroscopy (NIRS) can predict digestibility in real time, using chemometric algorithms, which quantitatively correlate feed and fecal spectral and analytical data (calibration equations). However, the accuracy of the predictions with the current methodology limits the possibilities of NIRS to generally rough estimates of digestibility. The present doctoral thesis, using the rabbit as a reference animal, aimed to improve the digestibility prediction accuracy in NIRS by: a) developing and evaluating a hybrid approach which combines fecal and feed spectral data, using an indigestible marker (phytogenic silica – PS) as a connecting link and b) evaluating the use of artificial neural networks (ANN) in the development of robust predictive models for predicting digestibility. Thus, the first hypothesis examined was whether PS could serve as an internal indigestible marker to calculate digestibility in rabbits by investigating its fecal recovery (to determine if it is 100% indigestible) and the ease of accurate quantification using simple and reliable methods in a very large number of samples and in a short time. The second hypothesis investigated was whether development of NIRS chemometric models for the quantification of PS along with major nutrients (XO, OO, AO, etc.) can be carried out in feed and feces to assess the applicability of a hybrid approach for calculating digestibility based on the chemical characteristics of feed and feces and excluding any physical information of the spectral data, which might negatively affect digestibility predictions. The examination also explored the possibility of using ANN to develop optimal digestibility prediction models using only the fecal spectral data (which also contain physical information). The third hypothesis examined in this thesis was whether and to what extent the modern features of NIRS can be utilized in order to decide if the proposed methodology can be applied in practical conditions to various types of samples (fresh, unground, dried, etc.). K). In this context, the use of artificial neural networks was again examined to investigate whether and to what extent they can improve the predictive power of NIRS models on unprepared fecal samples. To investigate the first hypothesis, thirty fecal samples from a previous study were used to calculate the coefficient of total tract apparent digestibility (CTTAD) of the dry matter (DM) and the gross energy (GE) of the same basic diet according to the European Reference Method (ERM), the AIA technique and the examined PS technique (n = 10 rabbits/method). PS is the major fraction of hydrochloric acid-insoluble ash (AIA), which is used as an internal marker in the determination of digestibility. However, compared to AIA, PS can be quantified by more rapid and sensitive methods. Therefore, the present experiment aimed to evaluate PS as an alternative marker to AIA for the calculation of digestibility in rabbits. The results showed that the concentrations of AIA in feed and feces were 12.89 ± 0.066 and 31.58 ± 0.042 g/kg, while those of PS were 13.98 ± 0.066 and 36.59 ± 0.059 g/kg, respectively (mean ± SEM). Fecal marker recovery was 100.3 ± 1.43 % and 100.8 ± 1.48 % for AIA and PS (P= 0.939) technique, respectively. The CTTAD of DM did not differ between methods (P= 0.935) and was 0.589 ± 0.0055, 0.591 ± 0.0056 and 0.588 ± 0.0060 for ERM, AIA and PS, respectively. Also, the CTTAD of GE was similar between all methods (P= 0.914) and was 0.581 ± 0.0058, 0.585 ± 0.0060, and 0.584 ± 0.0064 for ERM, AIA, and PS, respectively. PS determination required much less sample amount (by 98 %) and reagent volumes (by 68–82 %) and allowed the analysis of more samples (7-fold greater) compared to AIA. In conclusion, PS was evaluated as a reliable internal marker that offers the possibility to calculate the digestibility with the same accuracy as the reference method in rabbits. To examine the second and third hypotheses of this thesis, a total of 27 feed samples and 282 feecal samples from an in vivo digestibility experiment in rabbits were used. In addition, 43 feed samples were collected from commercial units to expand the data set. Each data set was randomly divided into calibration and validation sets. First, calibrations were developed to predict the chemical composition of feed and manure, including PS. The feed samples were examined in two forms (intact pellets, ground pellets), while the fecal samples were examined in three forms (fresh unground, dried unground and dried ground). Fecal spectra were used to predict the CTTAD of DM, organic matter (OM) and crude protein (CP). Calibrations were assessed a) internally by cross-validation using the leave-one-out method for feed and the k fold method for feces and b) externally with an independent set. In addition, forward propagation models were developed with artificial neural networks (ANNs) for the fecal samples in the three aforementioned forms in order to evaluate the prediction of the digestibility of DM, OM and CP using only the fecal spectral data. Most feed chemical parameters were accurately predicted (R2val > 0.8), but only in the ground samples. The analysis of the unground samples gave satisfactory results only for ΞO and OA (R2val = 0.80 and R2val = 0.88, respectively). Feces’ chemical parameters were accurately predicted, only for DM and CP (R2val = 0.89 and R2val = 0.88, respectively) in fresh feces. The analysis of dried ground fecal samples gave satisfactory results only for CP (R2val = 0.92). Regarding the PS marker, it was predicted in the feed accurately in the independent set (R2val = 0.75) and less accurately, but relatively satisfactorily in feces (R2val = 0.70). The prediction of nutrient digestibility using fecal spectra had a relatively good value (R2val > 0.75), but only for the dried and ground samples. On the contrary, the use of artificial neural networks (ANN) significantly improved the predictive power of the models both in fresh (intact fecal pellets) and dried intact fecal samples, which was particularly evident for the model predicting the digestibility of CP (R2val > 0 .91). The next step was to calculate the nutrient digestibility and then compare with the total collection method and indirect NIRS (where digestibility was predicted using the fecal spectral data alone). The NIRS models showed an accurate prediction of the PS index, and the comparison of the calculated digestibility coefficients with those of the reference method showed no significant differences. In conclusion, the hybrid approach can accurately estimate the digestibility of nutrients and furthermore, it only relies on the chemical information and not on any physical property of feed and feces, which may negatively affect digestibility prediction. In addition, the use of ANNs can improve the predictive power of digestibility models, especially in unprepared samples (fresh intact fecal pellets), which is of particular practical importance. The results of this thesis are followed by some limitations that must be taken into account. Initially, the models developed are for the specific samples used for calibration/validation, so using the models beyond these samples, i.e., on future unknown samples, may not give similar results. For the wide application of the models, an expansion of the calibration groups is also required. As observed in some of the examined parameters, the unsatisfactory results from the validation groups, despite the very good statistics in the calibration groups, indicated that the expansion of the sample dataset is required to increase the variability of the considered parameter.
Degree
thesis:*- Grantor dc:publisher
- Agricultural University of Athens
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Fortatos, Efstathios
- Φορτάτος, Ευστάθιος
Subjects
dc:subject × 17- Φασματοσκοπία ανάκλασης εγγύς υπερύθρου
- Διαιτητική αξία
- Ζωοτροφές
- Πεπτικότητα θρεπτικών συστατικών
- Κόνικλοι
- Τεχνητά νευρωνικά δικτύα - ΤΝΔ
- Νear infrared reflectance spectroscopy
- Feedstuff
- Nutritive value
- Digestibility
- Rabbits
- Artificial neural networks - ΑΝΝ
- Επιστήμη Ζωικής Παραγωγής
- Γεωπονικές Επιστήμες και Κτηνιατρική
- Animal and Dairy Science
- Agricultural and Veterinary Sciences
- Dairy and Animal Science
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/54812