Molecular factor computing for predictive spectroscopy

Purpose. The concept of molecular factor computing (MFC)-based predictive spectroscopy was demonstrated here with quantitative analysis of ethanol-in-water mixtures in a MFC-based prototype instrument. Methods. Molecular computing of vectors for transformation matrices enabled spectra to be represented in a desired coordinate system. New coordinate systems were selected to reduce the dimensionality of the spectral hyperspace and simplify the mechanical/electrical/computational construction of a new MFC spectrometer employing transmission MFC filters. A library search algorithm was developed to calculate the chemical constituents of the MFC filters. The prototype instrument was used to collect data from 39 ethanol-in-water mixtures (range 0-14%). For each sample, four different voltage outputs from the detector (forming two factor scores) were measured by using four different MFC filters. Twenty samples were used to calibrate the instrument and build a multivariate linear regression prediction model, and the remaining samples were used to validate the predictive ability of the model. Results. In engineering simulations, four MFC filters gave an adequate calibration model (r²=0.995, RMSEC=0.229%, RMSECV=0.339%, p=0.05 by f test). This result is slightly better than a corresponding PCR calibration model based on corrected transmission spectra (r²=0.993, RMSEC=0.359%, RMSECV=0.551%, p=0.05 by f test). The first actual MFC prototype gave an RMSECV=0.735%. Conclusion. MFC was a viable alternative to conventional spectrometry with the potential to be more simply implemented and more rapid and accurate.