CARNAC-D: a progress report

Introduction
Analysis by spectroscopic methods is based on the assumption that the spectroscopic properties of the sample arise from its composition. CARNAC is a method for using near infrared (NIR) spectroscopic data to predict composition via a database of similar samples which have been previously analysed for the analyte of interest. Conventionally, NIR data is used via the construction of a predictive model using techniques such as MLR, PLR, or PLS; CARNAC works by searching the database for samples which are very similar to the spectrum of the unknown sample. It then calculates the predicted analysis by a weighted average of the composition of these selected samples.
Experimental
The NIR data used is from publicly available data of pharmaceutical tablets. Calculations are made using a suite of MATLAB programs.
Results and discussions
The CARNAC procedure is a member of a group of methods known as “Local calibrations”; it was the first such method to be applied to NIR data. It requires large databases of analysed samples, fast computers with large memories and scanning spectrometers. None of these were widely available when it was first proposed and demonstrated in 1988 and the initial development received limited interest. In 2006 we published details of the re-development of the idea, CARNAC-D with some new methods programmed in MATLAB, when all the requirements were much more widely available. In 2006-7 we cooperated in a study of methods for analysis of animal feed and the results, reported in 2008, showed that CARNAC produced superior results compared to those produced by four other methods. We expect to publish additional excellent results from three databases in the near future.
The poster will describe the method and demonstrate its use with a pharmaceutical database which has been the subject of several investigations.