Construction of different calibration models by FTIR/ATR spectra and their application in screening of phenylketonuria

Objective: To construct different calibration models by using FTIR/ATR spectra and apply them for the screening of phenylketonuria for the newborns. Method: 69 dried blood spots samples were collected, of which the concentrations of phenylalanine (Phe) and tyrosine (Tyr) were determined by tandem mass spectrometry method. The FTIR/ATR method was used to collect the spectra of the samples. After various pretreatments with the original spectra, such as smoothing, derivative, vector normalization, Concave rubberband correction, the new spectra were fed into different methods to construct calibration models for the Phe concentrations of the samples. The kernel function, consensus strategy and their combination were integrated into the popular partial least squares method. The obtained models were then assessed by calculating coefficient of determination, root mean square error, mean relative error, etc. Result: After a careful comparison, the cKPLS model performed the best, and the result yielded were much more accurate and stable. Conclusion: The present cKPLS model, due to the combination of the kernel function and the consensus strategy, can be successfully applied to the prediction of Phe concentration for the phenylketonuria newborns. Also, we believe that the cKPLS method will be of help for FTIR/ATR spectra calibration problems in other complex systems. (C) 2017 Elsevier B.V. All rights reserved.