The data came from a national nutrition and health surveillance program covering four Chinese provinces. Of 12,567 children and adolescents initially assessed, 6,960 were included in the final analysis. Rather than treating protein as a single dietary variable, the researchers expressed each of 20 amino acids as a percentage of total protein intake, allowing them to ask whether the relative mix of amino acids tracked with fasting glucose status.

Elevated fasting glucose was the outcome of interest throughout the study. Feature selection was performed with the Boruta algorithm, and propensity score matching was used to reduce confounding by balancing covariates between comparison groups.

Conditional logistic regression was applied to estimate the relationship between each amino acid’s proportional intake and the risk of elevated fasting glucose, with odds ratios reported per 1-percentage-point increment in the proportion of total protein intake. This design means the reported effects describe shifts in amino acid composition within matched participants, not the consequences of adding or removing grams of protein from the diet. The machine learning component added a second, complementary lens.

The authors built predictive models using six algorithms: LightGBM, XGBoost, support vector machines, k-nearest neighbors, CatBoost, and naive Bayes. The best-performing model was LightGBM, which achieved an area under the curve of 0.831 when combining the 20 amino acids with 12 confounding factors, including fat intake, total protein intake, hypertension, and both general and central obesity. An AUC of 0.831 indicates reasonably strong discrimination between participants with and without elevated fasting glucose, though this is a modeling result within the study sample rather than a validated clinical screening tool.

To interpret what the model was relying on, the team applied SHAP, a technique that quantifies each feature’s contribution to individual predictions. When SHAP-based feature importance rankings were combined with the conditional logistic regression results, four amino acids emerged consistently. Methionine showed the strongest positive association, with an odds ratio of 2.03 (95% confidence interval: 1.61 to 2.57) per 1-percentage-point increase in its share of protein intake.

Threonine followed with an odds ratio of 1.59 (95% CI: 1.32 to 1.92). In the opposite direction, arginine had an odds ratio of 0.82 (95% CI: 0.74 to 0.92) and tyrosine an odds ratio of 0.71 (95% CI: 0.59 to 0.86). All four associations were statistically significant, with P values below 0.01.

Because amino acids do not act in isolation, the researchers also ran stratified analyses to check whether the associations held across subgroups of the study population. The directions of the associations were generally consistent across strata, suggesting that the methionine, threonine, arginine, and tyrosine signals were not driven by a single subgroup within the four provinces. The nonlinear dimension of the findings came from a generalized additive model, a flexible regression framework that can capture curved, rather than strictly linear, dose-response relationships.

In this analysis, methionine, arginine, and tyrosine all showed significant nonlinear relationships with elevated fasting glucose risk, with P values for nonlinearity below 0.05. In practical terms, this means the association between these amino acids and glucose status changed in shape across the range of intake proportions observed, rather than rising or falling at a constant rate. Threonine, by contrast, was highlighted in the linear regression and SHAP rankings but was not flagged for a significant nonlinear pattern in the GAM analysis.

The authors conclude that dietary amino acid intake is strongly associated with elevated fasting glucose levels in this population of children and adolescents, and that methionine, threonine, arginine, and tyrosine may collectively be important dietary correlates of abnormal glucose metabolism. Several limitations frame how these results should be read. The study is cross-sectional, so it captures associations at a single point in time and cannot establish that amino acid composition influences fasting glucose over time, and reverse causation is possible, since children with altered glucose metabolism may have different dietary patterns for reasons unrelated to the amino acids themselves. The analysis was restricted to participants from four Chinese provinces within one surveillance program, and the final sample of 6,960 was smaller than the 12,567 initially assessed, raising the possibility of selection effects. The machine learning results, including the LightGBM AUC, describe model performance within this dataset rather than validated predictions for new populations. The article’s final, formatted version had not yet been published at the time this summary was prepared. Within those boundaries, the study adds a compositional angle to pediatric nutrition research: what matters for glucose-related outcomes may include not only how much protein children eat, but which amino acids that protein delivers.

Subject of Research: Associations between dietary amino acid composition and elevated fasting glucose in children and adolescents

Article Title: Dietary amino acid composition and elevated fasting glucose in children and adolescents from four Chinese provinces: a cross-sectional study combining regression and machine learning

Article References: Chen, Y., Shang, Y., Wang, H., Zhu, Q., Liu, Y., Tian, M., Liu, C., Sun, Y., Tang, Z., & Yu, L. (2026). Dietary amino acid composition and elevated fasting glucose in children and adolescents from four Chinese provinces: a cross-sectional study combining regression and machine learning. Frontiers in Public Health, 14, Article 1912084. https://doi.org/10.3389/fpubh.2026.1912084

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1912084

Keywords: dietary amino acids, elevated fasting glucose, children and adolescents, machine learning, LightGBM, SHAP, methionine, threonine, arginine, tyrosine, cross-sectional study, glucose metabolism