Trusted information & resources for animal nutrition.

Technical Resources

Assessment of the accuracy of NIRS technology to determine the proximal composition and amino acid content of commercial soy-bean meal samples

Dardabou, L.,1, G. Mateos, L. Aguirre and M. Ibañez
2022

In total, 30 commercial samples of soybean meal (SBM) from USA (n = 11), Brazil (n = 10), and Argentina (n = 9) were collected randomly from European feed compound plants by trained personnel. The DM, CP, amino acids, ether extract after HCl hydrolysis (EEh), and CF were analyzed by wet chemistry (WCh) and NIRS. All wet analyses were performed in the same laboratory using official methods. NIRS values were determined by 4 specialized European laboratories (A, B, C, and D) using their own calibration models. Independently of the SBM origin, the average and range of WCh values (as fed bases) were 88.7 and 85.6 to 91.1% for DM, 46.2 and 44.2 to 49.8% for CP, 2.41 and 1.40 to 3.30% for EEh, 4.66 and 4.10 to 5.20% for CF, 2.91 and 2.75 to 3.06% for Lys, 0.63 and 0.53 to 0.68% for Met, and 0.67 and 0.59 to 0.75% for Cys. Bias, slope, and unexplained error were the statistics used to investigate the accuracy of WCh vs. NIRS technology, including comparisons among the 4 laboratories. The analysis was supported by a scatter plot of the ratio of NIRS to WCh (di) vs. NIRS values (ni) on a logarithmic scale, including bias confidence limits (BCL), limits of agreement, and any other trend that could be detected in the scatter plot. DM content was properly determined by the 4 laboratories, with 95% of the ratios within the 0.981 to 1.031. CP values were close to reference values, with a mean ratio between 0.992 and 1.048. The BCL variation among labs implied that the underestimation and the overestimation for the CP varied between 2% and 5%, respectively. For EEh and CF, the variability of the ratio ranged from 0.49 to 2.37 and from 0.47 to 1.29, respectively. The scatter plot trends showed that the magnitude of the error detected depended on the variable measured. For Lys, NIRS values were similar among companies, with laboratory “C” showing the lowest rate and an underestimation of around 7%, whereas for laboratory “D” the overestimation was as high as 7%. For Met, laboratory “C” had with a slightly higher prediction error than all the others, with overestimation and underestimation being of 6% and 4%, respectively. The error for Cys varied by 4% and 9%, for underestimation and overestimation, respectively. NIRS data from laboratories “”A”” and “”D”” were similar to WCh data with a SSEP of 0.02.

In summary, the data showed that NIRS technology is a valid alternative to WCh to estimate the chemical constituents of commercial SBM. However, its accuracy according to the analyzed component, the company that developed the calibration data, and factors, such as the country of origin of the SBM, that affects the chemi-cal composition of the samples.