Probabilistic latent semantic analysis of composite excitation-emission matrix fluorescence spectra of multicomponent...

In the present work, a simple and fast analytical procedure involving minimum user intervention was developed by combining the excitation-emission matrix fluorescence (EEMF) spectroscopy with Probabilistic latent semantic analysis (pLSA) technique. Akaike Information Criterion (AIC) was used to enable the user to automatically select the optimum model for analysing the mixtures of fluorescent components. The utility of the present work was successfully evaluated by analysing the dilute aqueous mixtures of certain fluorescent molecule such as Catechol, Hydroquinone, Indole, Tryptophan and Tyrosine of biological relevance. The developed AIC assisted pLSA model of five components explained >90% variance of spectral data sets. The identity between the pLSA retrieved spectral profiles was established using similarity index (SI) parameter in automatic manner. The SI values were found to be close to unit values for each of the five analyzed molecules. The regression parameter between the actual and pLSA predicted concentrations were found to be well within acceptable limits. Both root mean square of calibration and predictions for each of the five fluorescent molecules were found to be <1%, whereas, the square of the correlation coefficient (R²) value was found to be >0.98 suggesting the developed pLSA model was quite precise in analysing both calibration and validation set samples. The uniqueness of the developed pLSA model for EEMF spectroscopic data was successfully tested using the sequential quadratic programming (SQP) algorithm. The differences between the upper and lower bands in SQP were found to be ≤0.005. In summary, the proposed approach serve as swift and simple analytical tool for the analysis of fluorescent mixtures without involving pre-separation step.