Efficient spectroscopic calibration using net analyte signal and pure component projection methods

摘要

In the wake of FDA's finalisation of the process analytical technology guidance to industry, the application of near infrared (NIR) spectroscopy for quality analysis in pharmaceutical manufacturing has continued to grow. The required level of variation needed to develop a NIR method often exceeds that observed in a well-controlled pharmaceutical production process. This insufficiency can be addressed by developing non-production samples to introduce range, but at high cost in labour and complexity. The recently-introduced pure-component projection (PCP) method utilises the information in the spectral characteristics of pure sample constituents to reduce NIR spectra to a univariate signal, thereby mitigating the need for non-production samples. The PCP method is compared to net analyte signal (NAS) processing and PLS regression calibration when relatively little calibration data are available. The predictive performance of all algorithms was observed to be similar, although NAS and PCP have advantages in selecting the optimal number of latent variables for calibration. PCP holds a definite advantage as the only algorithm capable of producing a sensitive, linear regression coefficient vector without chemical reference data or non-production samples.