Correlation coefficient optimization in partial least-squares regression with application to ATR-FTIR spectroscopic analysis

Abstract

A wavelength selection method for spectroscopic analysis, named correlation coefficient optimization coupled with partial least-squares (CCO-PLS), is proposed, and was successfully employed for reagent-free ATR-FTIR spectroscopic analysis of albumin (ALB) and globulin (GLB) in human serum. By varying the upper bound of correlation coefficient between absorbance and analyte's content, the CCO-PLS method achieved multi-band selection. Two PLS-based methods, which used a waveband having positive peaks of the first loading vector (FLV) and a combination of positive peaks of the correlation coefficient spectrum, were also conducted for comparison. Based on the leave-one-out cross-validation for CCO-PLS, appropriate waveband combinations for ALB and GLB were selected, the root-mean-square errors of prediction for validation samples were 1.36 and 1.35 (g L⁻¹) for ALB and GLB, respectively, which were better than the two comparison methods. The CCO-PLS method provided a new approach for multi-band selection to achieve high analytical accuracy for molecular absorption bands that were composed of several spaced wavebands.