Chemical Composition, Nutritional Quality and Yield in Lactuca aff. indica Varieties Using Multivariate Analysis Techniques

Abstract

Multivariate analysis techniques can be useful for analyzing data that seeks to separate food plant cultivars according to yield, leaf quality and nutritional value. Thus, we used, validated and compared principal component analysis (PCA), Kohonen’s organizable maps (SOM), a nonsupervised competitive learning artificial neural network formed by a grid of (artificial neurons) and multifactorial analysis to differentiate three cultivars of tree lettuce (Lactuca aff indica), an unconventional food plant (PANC). For the differentiation, physicochemical variables of the leaves were evaluated in order to determine their sensory quality (total titratable acidity, total soluble solids, pH), antioxidant properties (vitamin C and total phenolics), nutritional and mineral composition (centesimal composition and macro and microelements), and yield results indicate that the highest leaf yield occurred in the first cutting for the three varieties evaluated, the nutritional quality of the leaves increased progressively as the experiment progressed, reaching significantly higher values in the third cut for some macro and micronutrients: Ca (278.9 g kg–1), Mg (64.1 9 g kg–1), S (2.53 g kg–1), Fe (1.78 g kg–1), Mn (1.93 mg 100 g–1). Among the varieties evaluated, SG stood out for presenting the best antioxidant properties, due to the highest concentration of ascorbic acid (355.18 mg 100 g–1) and total phenolics (569.2 mg GAE 100 g–1) in the second cut, along with a significantly higher average of free radical scavenging (SRL = 17.4%) when compared to PP and PR with values of 14.0 and 9.1%, respectively. According to the validation of the multivariate methods, all were suitable for analyzing the data obtained. SOM and PCA showed similar results, but PCA needed three components to be able to explain 75% of the data in a three-dimensional graph, while SOM made it possible to more efficiently explore the tendency to group the samples into seven groups according to the similarities and differences in the variables evaluated. This is due to its ability to reduce the size of the data and maintain a true representation of the relevant properties of the input vectors, generating a two-dimensional map that allows the results to be easily visualized.