Scholarlink

Background <p>Polygenic predictors can enhance screening for metabolism-related traits such as body mass index (BMI) and type 2 diabetes (T2D). However, these predictors explain limited phenotypic variance and face implementation challenges in non-European populations. Dietary patterns are well-established metabolic risk factors that remain under-investigated in quantitative risk stratification models.</p> Methods <p>We developed and evaluated risk stratification models combining polygenic predictors and data-driven dietary scores (DDS) for BMI and T2D in 14,346 Native Hawaiians from the Multiethnic Cohort (MEC-NH), a population with high prevalence of obesity and T2D. Using 5,374 participants with genetic data, we integrated publicly available large-scale GWAS summary statistics to develop cross-ancestry polygenic score (PGS) models using PRS-CSx. We developed DDS using machine learning algorithms on 520 dietary variables and evaluated model performance in held-out test sets.</p> Results <p>Trans-ancestry PGS achieved better prediction accuracy than single-ancestry models for both phenotypes (partial R² [SE] = 0.12 [0.04] vs. 0.03–0.09 for BMI; liability R² [SE] = 0.09 [0.04] vs. 0.01–0.07 for T2D). The best-performing DDS was based on a Random Forest model and substantially explained BMI variation (partial R² [SE] = 0.12 [0.01]), comparable to genetic prediction. Combined models integrating PGS and DDS significantly outperformed single-predictor models for BMI (adjusted R² = 0.29 vs. 0.21, <i>P</i> &lt; 10⁻¹¹⁷). For T2D, combined models showed marginal but significant improvement over PGS alone. The BMI dietary score additionally associated with multiple chronic diseases, with effects partially mediated through inflammatory and lipid pathways.</p> Conclusion <p>Trans-ancestry polygenic scores and data-driven dietary scores provide complementary information for metabolic trait prediction. Combined genetic-dietary models significantly outperform single-predictor approaches, with improvement being most pronounced for BMI. In Native Hawaiians, systematic integration of dietary information substantially improved BMI prediction, demonstrating the value of incorporating modifiable environmental factors alongside genetic information.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Incorporating dietary information to enhance polygenic prediction models with applications to body mass index and type 2 diabetes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月18日

查看全文
  • Eunice Y. Lee,
  • Bryan L. Dinh,
  • Ji Tang,
  • Samantha A. Streicher,
  • Xinran Wang,
  • Subarna Biswas,
  • He Tian,
  • Xian Yu,
  • Kekoa Taparra,
  • Take Naseri,
  • Satupa’itea Viali,
  • Daniel E. Weeks,
  • Jenna C. Carlson,
  • Christopher A. Haiman,
  • Loïc Le Marchand,
  • Gertraud Maskarinec,
  • Lynne R. Wilkens,
  • Song-Yi Park,
  • Charleston W. K. Chiang

摘要

Background

Polygenic predictors can enhance screening for metabolism-related traits such as body mass index (BMI) and type 2 diabetes (T2D). However, these predictors explain limited phenotypic variance and face implementation challenges in non-European populations. Dietary patterns are well-established metabolic risk factors that remain under-investigated in quantitative risk stratification models.

Methods

We developed and evaluated risk stratification models combining polygenic predictors and data-driven dietary scores (DDS) for BMI and T2D in 14,346 Native Hawaiians from the Multiethnic Cohort (MEC-NH), a population with high prevalence of obesity and T2D. Using 5,374 participants with genetic data, we integrated publicly available large-scale GWAS summary statistics to develop cross-ancestry polygenic score (PGS) models using PRS-CSx. We developed DDS using machine learning algorithms on 520 dietary variables and evaluated model performance in held-out test sets.

Results

Trans-ancestry PGS achieved better prediction accuracy than single-ancestry models for both phenotypes (partial R² [SE] = 0.12 [0.04] vs. 0.03–0.09 for BMI; liability R² [SE] = 0.09 [0.04] vs. 0.01–0.07 for T2D). The best-performing DDS was based on a Random Forest model and substantially explained BMI variation (partial R² [SE] = 0.12 [0.01]), comparable to genetic prediction. Combined models integrating PGS and DDS significantly outperformed single-predictor models for BMI (adjusted R² = 0.29 vs. 0.21, P < 10⁻¹¹⁷). For T2D, combined models showed marginal but significant improvement over PGS alone. The BMI dietary score additionally associated with multiple chronic diseases, with effects partially mediated through inflammatory and lipid pathways.

Conclusion

Trans-ancestry polygenic scores and data-driven dietary scores provide complementary information for metabolic trait prediction. Combined genetic-dietary models significantly outperform single-predictor approaches, with improvement being most pronounced for BMI. In Native Hawaiians, systematic integration of dietary information substantially improved BMI prediction, demonstrating the value of incorporating modifiable environmental factors alongside genetic information.