We have investigated the use of principal component analysis (PCA) to describe and assess mid-infrared spectral data obtained from complex biological samples containing sucrose, fructose, and glucose. The correlation coefficients between spectral data and chemical values of each variable (sucrose, glucose, fructose, total sugars, and reducing sugars) showed that in each case, axes 1, 3, 4, and 5 had the highest values. These values also indicated which axes each variable was mostly correlated with. The results also showed that the samples were distributed according to their sucrose concentrations (or total sugars) along a concentration gradient in the projection plan formed between axes 1 and 3. No clear discrimination according to concentration was observed with other factorial maps. Prediction equations that linked sucrose, fructose, glucose, total sugar, and reducing sugars concentrations to the spectral data were established by regression on the principal component. Very high correlation coefficients values between the first 10 axes and the chemical values were obtained (between 0.9757 and 0.998). From such aqueous biological samples containing a ternary mixture of sucrose, fructose, and glucose, it was possible to (1) identify the characteristic IR bands of these different sugars (and their combination: reducing sugars/total sugars) and (2) to specifically measure their concentrations with a relatively good accuracy.
Predicting bovine tuberculosis status of dairy cows from mid-infrared spectral data of milk using deep learning
