image: Schematic illustration of an AI‑glycomics framework for precision nutrition. Multi‑source information—including polysaccharide structures, gut microbiota, host phenotypes, and database resources—is integrated for AI interpretation and machine‑learning modeling. The resulting polysaccharide structure–microbiome response–metabolic output models support function‑oriented formulation design, including personalized nutrition plans, smart capsules, nanocarriers, and hydrogels. Intervention outcomes are subsequently fed back into the system, forming a closed‑loop framework for precision nutritional intervention and iterative optimization.
Credit: Food & Medicine Homology, Tsinghua University Press
Polysaccharides represent one of the most structurally diverse groups of biological macromolecules. Their biological and physicochemical properties are jointly governed by many interwoven structural features: molecular‑weight distribution, monosaccharide composition, glycosidic linkages, branching topology, chemical modifications, higher‑order conformation and aggregation states. This high structural complexity makes polysaccharide structure‑activity relationships extremely difficult to decode with conventional empirical and single‑factor research strategies.
A cross‑institutional research team lays out an AI‑powered roadmap to transform polysaccharide research in a Perspective article published in Food & Medicine Homology on August 19, 2026.
“Traditional polysaccharide research heavily depends on empirical testing. Artificial intelligence opens new possibilities to move the field from experience‑driven discovery to data‑driven prediction, from fragmented analytical workflows toward end‑to‑end integration, and from passive structural characterization toward function‑oriented rational design,” said Jia‑Lun Duan, corresponding author from Nanjing University of Chinese Medicine.
The team argues that algorithm sophistication is not the main limiting factor for AI‑driven polysaccharide research at present. The real bottleneck lies in the shortage of high‑quality, standardized, comparable and traceable datasets. To address this gap, the researchers put forward a minimum‑information framework for “AI‑ready polysaccharide record”. This standard covers biological sources, extraction and processing history, composition and purity, primary‑to‑higher‑order structural information, raw analytical evidence, bioactivity measurements, model provenance, validation records and prediction uncertainty. Such a complete sample‑structure‑assay chain can build solid data foundations for reproducible AI models.
The Perspective reviews multiple emerging digital representation approaches for polysaccharides. These range from conventional feature vectors and graph‑based representations of branched carbohydrate architectures, to atom‑level heterogeneous graphs and direct learning from raw NMR, MS and IR spectral data. Multimodal AI systems are capable of fusing structural descriptors, spectroscopy outputs, literature knowledge, molecular simulation results and multi‑omics datasets into unified computational representations.
AI is also reshaping polysaccharide structural elucidation. Multimodal machine‑learning models can combine complementary signals from multiple spectroscopic techniques. Generative AI enables an inverse “hypothesis‑validation” loop: candidate polysaccharide structures are generated, theoretical spectra are predicted, and outputs are cross‑checked against real‑world experimental measurements. This workflow promises higher throughput and better reliability for structural identification.
Beyond characterization, AI supports prediction of higher‑order conformations, physicochemical properties and structure‑activity links, and helps pinpoint key structural motifs responsible for bioactivity as well as polysaccharide‑protein interactions at atomic scale. The authors stress that explainable artificial intelligence and causal inference are indispensable, to separate true causal structural determinants from spurious correlations introduced by experimental confounders.
Looking further ahead, generative AI enables function‑driven de novo polysaccharide design. Rather than following the classic “extraction‑characterization‑validation” pipeline, researchers can start directly from target functional requirements, such as immunomodulation or prebiotic effects, and computationally generate candidate polysaccharide structures with predefined molecular‑weight, linkage, branching and modification profiles. Synthetic biology and high‑throughput experiments then convert these digital designs into real molecules. Results flow back to computational models to sustain an iterative “design‑build‑test‑learn” closed‑loop cycle.
The team also connects AI glycomics to food‑medicine homology and precision nutrition. Integrated datasets combining polysaccharide structures, gut microbiome profiles, metabolomics, host phenotypes and clinical data allow researchers to model the “glycan structure‑microbiome‑metabolite‑health outcome” axis. The concept of a Digital Twin Gut is introduced: a personalized simulated intestinal environment that simulates human responses to polysaccharide‑based interventions before real‑world dietary or clinical application.
The authors outline phased development priorities for the community. Near‑term goals include reaching consensus on metadata standards, physical reference materials and benchmark datasets. Multicenter datasets and active‑learning studies will support robust cross‑laboratory validation. In the long run, fully closed‑loop platforms will realize rational development of functional polysaccharides.
“AI’s greatest value is not simply accelerating existing experiments. It empowers scientists to uncover hidden structural rules and biological mechanisms hard to detect via conventional approaches. Still, AI models must always be tightly coupled with experimental validation and expert domain knowledge,” explained Wen‑Yi Kang, corresponding author at Henan University. Deep integration between standardized data, interpretable AI, causal reasoning, synthetic biology and wet‑lab experiments will unlock a new era for rational design and precision application of functional polysaccharides.
Other contributors include Shi‑Kang Chen from Henan University and Tao Li from Qilin Heheng Network Technology INC.
This work was supported by relevant research grants hosted by the National R&D Center for Edible Fungus Processing Technology.
DOI Link:
https://doi.org/10.26599/FMH.2026.9420166
Article Title
Functional reinterpretation of food‑medicine homologous polysaccharides: From bioactive macromolecules to multifunctional immunological materials
Article Publication Date
19-Aug-2026