Image-based modeling in the selection of cowpea genotypes for aluminum tolerance

Abstract

In acidic soils, one of the main factors limiting crop yield is the presence of aluminum. However, some species may exhibit tolerance and mitigate the toxic effects of this element. This study aimed to evaluate and compare machine learning algorithms for the predictive analysis of aluminum ion accumulation sites, using the Evans blue dye to assess its interaction with root morphology, in cowpea genotypes. A completely randomized design, with five aluminum concentrations (0, 50, 100, 150 and 200 µM of AlCl₃·6H₂O) and fourteen cowpea genotypes, was used. Significant variations in aluminum tolerance were observed among the genotypes, with MNC06-908-39 and MNC06-895-1 being classified as the most sensitive, whereas MNC06-907-29 and MNC06-901-14 exhibited a higher tolerance. CB-27 was identified as the most tolerant overall, whereas MNC06-909-54, MNC06-909-55, MNC06-907-35, MNC06-908-39 and MNC06-909-76 were considered the highest sensitive ones. The quantitative dye analysis corroborated the qualitative findings, confirming the higher aluminum tolerance of the MNC06-907-29 and MNC06-901-14 genotypes, whereas MNC06-908-39 and MNC06-895-1 were the most sensitive ones. A predictive analyses, using the Random Forest, Decision Tree, k-Nearest Neighbors and Neural Network algorithms, demonstrated a clear genetic variability among the genotypes in response to aluminum stress, with the Neural Network showing the highest predictive accuracy.

KEYWORDS: Vigna unguiculata L., abiotic stress, machine learning.