Arusha Bashyal

The study uses multivariate and neural network (ANN) to characterize proximate composition and mineral content of 46 Nepalese cereal and legume germplasms. Kruskal Wallis showed differences (p < 0.05) among items. Principal Component Analysis (PCA) of the composition matrix demonstrated 80.77% of variance in three components (PC1 = 60.13%, PC2 = 12.25%, PC3 = 8.39%); PC1 distinguished mineral-rich legumes from carbohydrate-dominant cereals, while PC2 differentiated calcium- and phosphorus-rich millets. Ward's cluster analysis (HCA) identified four clusters: legumes, rice and sorghum, millets and buckwheat, and other cereals. ANN classifier, optimized over 25 hyperparameter combinations (size ∈ {2, 3, 5, 7, 10} × decay ∈ {0.0001–0.5}), achieved 74.5% LOOCV accuracy (κ = 0.714; macro-F1 = 0.851) under 10-fold cross-validation. The random forest classifier outperformed all others (100.0 %). Olden's analysis identified Iron, Sodium, and Phosphorus as principal discriminating variables. t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) confirmed the four-cluster structure, with mean AUC of 0.990. Thus, this study provides a foundation for Nepalese food database and dietary plannin

The study aims to use multivariate and artificial neural network (ANN) to characterize the proximate composition and mineral content of 46 Nepalese cereal and legume germplasms. ANOVA showed significant differences (p < 0.05) among the food items for all parameters. Principal Component Analysis (PCA) of the scaled composition matrix demonstrated 90.5% of the total variance in three components; PC1 distinguished mineral-rich legumes from carbohydrate-dominant cereals, while PC2 further differentiated calcium- and phosphorus-rich millets. Ward's hierarchical cluster analysis (HCA) identified four nutritionally distinct clusters namely legumes, rice and sorghum, minor millets and buckwheat, and other cereals. ANN classifier, optimized through a grid search over 25 hyperparameter combinations (size ∈ {2, 3, 5, 7, 10} × decay ∈ {0.0001–0.5}), achieved a classification accuracy of 97.0% (kappa = 0.964) under repeated 10-fold cross-validation. The random forest classifier outperformed all others (100.0 %). Olden's connection weight analysis identified potassium, protein, and calcium as the three principal discriminating variables. t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction independently confirmed the four-cluster structure, with a mean ROC-AUC of 1.000 for all. Thus, this study provides a validated analytical and methodological foundation for Nepalese food database and evidence-based dietary planning.

The study characterized physicochemical and functional quality diversity of 201 wheat (Triticum aestivum L.) genotypes from Nepal’s national wheat improvement programme using chemometric multivariate methods. A total of 603 observations were analyzed for 14 quality parameters, including proximate composition (moisture, crude protein, crude fat, crude fibre, total ash, carbohydrate), mineral content (iron, phosphorus, calcium), and functional properties (dry gluten, swelling capacity, gelation temperature, water absorption capacity, oil absorption capacity). Principal component analysis (PCA), k-means clustering, and partial least squares discriminant analysis (PLS-DA) with variable importance in projection (VIP) scoring were conducted in R. All variables differed significantly across four quality clusters (Kruskal-Wallis χ²(3) = 38.7–372.4; p < 0.001). Three principal components explained 74.47% of variance: PC1 (38.43%) represented protein-gluten quality; PC2 (22.40%) balanced mineral composition versus fat-starch interaction; PC3 (13.64%) reflected dietary fibre and mineral retention. K-means clustering identified four clusters: Cluster 1 (high-phosphorus mineral-dense), Cluster 2 (high-protein-gluten), Cluster 3 (iron-rich functional), and Cluster 4 (high-fat-fibre balanced). PLS-DA highlighted dry gluten (VIP=1.369), crude protein (VIP=1.300), and oil absorption capacity (VIP=1.272) as key discriminators. Nepal’s wheat germplasm exhibits four distinct quality clusters with implications for bread-making, mineral biofortification, and functional food development.