Net analyte signal calculation in multivariate calibration

Net analyte signal plays an important role in the calculation of figures of merit for characterizing a calibration model. Until now, its computation has only been feasible for the direct calibration model, which requires knowledge of pure spectra or concentrations of all contributing species in the calibration samples. An increasingly important calibration model is the inverse calibration model, which also allows for quantitation if the knowledge about the interferents is incomplete. This paper shows that net analyte signal computation is possible for the inverse calibration case. Application to the determination of protein content in wheat samples by near-infrared spectrometry is presented. Net analyte signal calculation was used to estimate selectivities (ratio of signal available for quantitation to the total measured signal). The selectivities were found to range between 0 and 2% of the measured reflectance signal. A new measure for outlier diagnosis based on the correlation of the net analyte signal to the regression coefficients vector is introduced and tested on the same data.