Versatile machine learning algorithms for FTIR spectroscopy: differentiating crosslinked and non-crosslinked gelatin samples

From the food and pharmaceutical industries there is a growing demand for biomaterials from industrial waste, which are biodegradable, non-toxic, non-carcinogenic and biocompatible, as is the case of the gelatin. This is a mixture of amino acid polymers with a variable molecular weight and a diversity of secondary structures (α, β and γ chains). The functional and mechanical properties of the gelatin depend on the chemical and structural characteristics. To improve these properties, crosslinking methodologies have been implemented, which form covalent and/or non-covalent interactions within the gelatin structure. Fourier transform infrared spectroscopy (FTIR) has been presented as an alternative to evaluate structural differences and monitor the cross-linking process of four types of gelatins. A corresponds to a standard gelatin with a high level of cross-linking, B, C and D are pharmaceutical gelatins with different degrees of cross-linking used in the formulation of soft capsules. For the FTIR results, machine learning classification models developed in the Python language were used to distinguish between cross-linked and non-cross-linked gelatin samples and two dimensionality reduction techniques (PCA, PLS) and four classification models (NCA-KNN, SVM, LDA, DT) were incorporated, all effectively classifying spectra across gelatin types in adjustment, training, test stages, and predictions, with higher precision observed for gelatins A, C, and D. FTIR analysis, being rapid, efficient, non-destructive and adaptable for in-line operations, has potential for early warning systems in cross-linking processes to avoid further denaturation of gelatin.