Use of random forest analysis to quantify the importance of the structural characteristics of beta-glucans for prebiotic development

Ka Lung Lam, Wai Yin Cheng, Yuting Su, Xiaojie Li, Xiyang Wu, Ka Hing Wong, Hoi Shan Kwan, Peter Chi Keung Cheung

Research output: Journal article publicationJournal articleAcademic researchpeer-review

5 Citations (Scopus)

Abstract

Beta-glucans as food hydrocolloids have demonstrated prebiotic potential recently. Novel hydrocolloid-based prebiotics could be developed by structural manipulation of existing carbohydrate candidates using different physical, chemical and/or biological approaches. However, there are no guidelines on choosing the most promising structural feature as the starting point(s) for modification. We performed a parallel in vitro fermentation of six native and carboxymethylated beta-glucans from mushroom, bacteria, seaweed and cereal against five probiotic bacteria, including bifidobacteria and lactobacillus, and determined their growth parameters including the lag phase, maximum growth rate, and maximum population increase. An unsupervised analysis pipeline was developed to elucidate the structure-property relationship of beta-glucans as prebiotic candidates and identified several chemical composition and structural features for modification to maximize their prebiotic potential. We adopted the random forest algorithm to quantify the relative importance of different composition and structural features in altering the three probiotic growth parameters. In terms of structural features, the desirable characteristics for prebiotic development are, in descending order of importance, low molecular weight, high water solubility, and no introduction of carboxymethyl functional group. Apart from structural features, we have identified the chemical composition as high purity, plays a significant importance in increasing the probiotic growth parameters, raising the relevance of including the purity in discussing the developmental cycle of prebiotic potential maximization. We suggest an integrated data alchemy approach including both machine learning and domain knowledge in making decision of choosing the promising starting point(s) and direction for modification of existing hydrocolloids in prebiotic research.

Original languageEnglish
Article number106001
JournalFood Hydrocolloids
Volume108
DOIs
Publication statusE-pub ahead of print - 7 May 2020

Keywords

  • Chemical composition
  • Cyclic prebiotic potential maximization
  • Data alchemy approach
  • Machine learning
  • Structural characterization
  • Structure-property relationship

ASJC Scopus subject areas

  • Food Science
  • Chemistry(all)
  • Chemical Engineering(all)

Cite this