Related Experiment Videos
A dataset-calibrated multi-module model for evaluating plant oil similarity to human milk lipids: single oil ranking,
Yuankai Wang1, Fanlong Wang1, Lu Wang1
1College of Food Science, Southwest University, No. 2, Tiansheng Road, Beibei, Chongqing, 400715, PR China; China-Hungary Belt and Road Joint Laboratory on Food Science, Chongqing, 400715, PR China; Chinese-Hungarian Cooperative Research Centre for Food Science, Chongqing, 400715, PR China.
Abstract:
Designing infant-formula lipids requires matching human milk across multiple lipid dimensions, but plant-oil screening lacks a unified objective method. We developed a model to assess similarity and optimize blends. Liposoluble bioactives, phospholipids, and triacylglycerol fatty acids of 20 plant oils were compared with lipid centroids of colostrum, transitional milk, and mature milk. Aitchison distances were converted to module similarities by quantile-logistic calibration and absolute-composition correction and integrated using jackknife-optimized weights. Corylus heterophylla Fisch. ex Trautv. ranked highest in colostrum, whereas Prunus dulcis (Mill.) D.A.Webb performed better in transitional and mature milk, with leading oils remaining stable across stages. Blend optimization showed Glycine max (L.) Merr. dominated phospholipid matching, while Prunus dulcis (Mill.) D.A.Webb, Idesia polycarpa Maxim., and Pistacia chinensis Bunge complemented fatty acid and integrated blends. Robustness analyses supported ranking stability, while PERMANOVA, SIMPER, and machine learning characterized compositional and methodological differences, supporting preliminary, dataset-specific feedstock screening.