Paper reading (三十二)：Personalized Nutrition by Prediction of Glycemic Responses（Results）

最新推荐文章于 2019-11-27 22:06:45 发布

盲人骑瞎马5555

最新推荐文章于 2019-11-27 22:06:45 发布

阅读量455

点赞数

分类专栏： Paper Reading 文章标签： personalized nutrition

本文链接：https://blog.csdn.net/wxw060709/article/details/102537197

版权

Paper Reading 专栏收录该内容

133 篇文章 9 订阅

订阅专栏

正文部分内容摘录：

2. Results

2.1 Measurements of postprandial responses, clinical data, and gut microbiome

human cohort关键信息：800 individuals；aged 18–70；not previously diagnosed with TIIDM
The cohort is representative of the adult non-diabetic Israeli population： with 54% overweight (BMI ≥ 25 kg/m2) and 22% obese (BMI ≥ 30 kg/m2) 保持与2014年以色列总体人口统计的比例相同，就可以类同于抽样。
These properties are also characteristic of the Western adult non-diabetic population，这个的参考来源于2008年世界卫生组织。为什么不是2014年呢？
figure 1 A中的饮食是主要是面包，黄油面包，葡萄糖等，这些食物是如何选择的呢？each consisting of 50 g of available carbohydrates
validataion cohort: 100 participants; dietary intervention: 26 participants
glycated hemoglobin (HbA1c%): prediabetes (HbA1c% ≥ 5.7%) and TIIDM (≥6.5%)
continuous glucose monitor (CGM)
previous studies found no significant differences between PPGRs extracted from CGMs and those obtained from either venous or capillary blood (Vrolix and Mensink, 2010) 又是一个铺垫性的研究
参加实验者被要求除了第一餐由实验组织人员提供标准餐以外，其他都跟他们平时的饮食相似。
a comprehensive profile：food frequency, lifestyle, and medical background questionnaires; anthropometric measures (e.g., height, hip circumference); a panel of blood tests; and a single stool sample, used for microbiota profiling by both 16S rRNA and metagenomic sequencing.
这部分主要对于participates进行了各种分析，证明cohort的选取是合理的。然后又介绍了采集cohort中每个人的数据的方法，并且对采集到的数据进行了初步的统计分析。

2.2 Postprandial glycemic responses associate with multiple risk factors

PPGRs与已知的很多risk factors强相关，如BMI (R = 0.24, p < $10^{-10}$ ), glycated hemoglobin糖化血红蛋白 (HbA1c%, R = 0.49, p < $10^{-10}$ ), wakeup glucose空腹血糖 (R = 0.47, p < $10^{-10}$ ), and age (R = 0.42, p < $10^{-10}$ )
These results suggest that the entire range of glucose levels of an individual may have clinical relevance, with different percentiles being more relevant for particular risk factors.

2.3 High interpersonal variability in the postprandial response to identical meals

从Figure 1A中可以看出，标准配餐的bread、bread&butter、glocuse三种类型每周重复两次。这样的实验设计正是为这部分研究做准备的。
high interpersonal variability, with the PPGRs of every meal type (except fructose) spanning the entire range of PPGRs measured in our cohort
The large interpersonal differences in PPGRs are also evident in that the type of meal that induced the highest PPGR differs across participants and that different participants might have opposite PPGRs to pairs of different standardized meals
For foods with a published glycemic index, our population-average PPGRs agreed with published values (R = 0.69, p < 0.0005), further supporting our data

2.4 Postprandial variability is associated with clinical and microbiome profiles

Notably, the TIIDM and metabolic syndrome risk factors HbA1c%, BMI, systolic blood pressure, and alanine aminotransferase (ALT) activity are all positively associated with PPGRs to all types of standardized meals, reinforcing the medical relevance of PPGRs.
In most standardized meals, PPGRs also exhibit a positive correlation with CRP, whose levels rise in response to inflammation
RAs of Actinobacteria are positively associated with the PPGR to both glucose and bread
The KEGG pathway of ABC transporters also exhibits positive association with several standardized meal PPGRs.
Several bacterial secretion systems are positively associated with most standardized meal PPGRs.
KEGG modules for transport of the positively charged amino acids lysine and arginine are associated with high PPGR to standardized foods, while transport of the negatively charged amino acid glutamate is associated with low PPGRs to these foods.
Taken together, these results show that PPGRs vary greatly across different people and associate with multiple person-specific clinical and microbiome factors.

2.5 Prediction of personalized postprandial glycemic responses

two-phase approach:

discovery phase. on the main cohort of 800 participants, performance was evaluated using a standard leave-one-out cross validation scheme 类似于训练阶段
validation phase，类似于验证阶段

features分为personal features和meal features（Time, nutrients, prev, exercise）
算法相关：boosted decision trees，cross-validation，leave-one-person-out？，
model主要参考：gradient boosting regression（reference：Greedy Function Approximation: A Gradient Boosting Machine）
特征池： a pool of 137 features representing meal content
特征总共分6类：meal content ; daily activity ; blood parameters; CGM-derived features; questionnaires; and microbiome features
baseline reference：“carbohydrate counting” model, as it is the current gold standard for predicting PPGRs（reference：4. Foundations of Care: Education, Nutrition, Physical Activity, Smoking Cessation, Psychosocial Care, and Immunization；Improving the Estimation of Mealtime Insulin Dose in Adults With Type 1 Diabetes）
This correlation approaches the presumed upper bound limit set by the 0.71–0.77 correlation that we observed between the PPGR of the same person to two replicates of the same standardized meal.
所以说这一部分阐述了所用到的model以及baseline model。

2.6 validation of personalized postprandial glycemic responses predictions on an independent cohort

Data from this additional cohort were not available to us while developing the algorithm.
No significant differences were found between the main and validation cohorts in key parameters, including age, BMI, non-fasting total and HDL cholesterol, and HbA1c%
R = 0.68 and R = 0.70 on the main and validation cohorts 重点看看这个R是什么

2.7 Factors underlying personalized predictions

这篇paper一共有14个大figure。。。每个figure由多个小图组成，总数为67。。。我不得不说，每个figure配色都很好看。
partial dependence plots (PDP)：used to study functional relations between features used in predictors
PDP有一些缺陷：it may be misleading due to higher-order interactions，但是在large datasets也还是经常会用。
the PDP of carbohydrates (Figure 4A) shows that as the meal carbohydrate content increases, our algorithm predicts, on average, a higher PPGR.
carbohydrate sensitivity is also person specific.
adding fat to meals may reduce the PPGR，the effect of fat varies across people.
The 72 PDPs of the microbiome-based features used in our predictor were either beneficial (21 factors), non-beneficial (28), or non-decisive (23) in that they mostly decreased, increased, or neither, as a function of the microbiome feature.
This may reflect limitations of the PDP analysis or result from a more complex relationship between these features, obesity, and PPGRs.
We found 20 statistically significant correlations (p < 0.05, FDR corrected) where microbiome factors termed non-beneficial correlated with risk factors, and those termed beneficial exhibited an anti-correlation.
These results suggest that PPGRs are associated with multiple and diverse factors, including factors unrelated to meal content.
我只想说千丝万缕。。。好复杂。。。

2.8 Personally tailored dietary interventions improve postprandial responses

a two-arm blinded randomized controlled trial and recruited 26 new participants
A clinical dietitian met each participant and compiled 4–6 distinct isocaloric options for each type of meal (breakfast, lunch, dinner, and up to two intermediate meals)
Participants then underwent the same 1-week profiling of our main 800-person cohort (except that they consumed the meals compiled by the dietitian)
providing the inputs (microbiome, blood parameters, CGM, etc.) that our algorithm needs for predicting their PPGRs.
Participants were then blindly assigned to one of two arms: prediction arm and expert arm, 两组的区别就在于前者是算法推荐的好的食物组合和坏的食物组合，而后者是专家推荐的，后者是对照组。
the predictor-based approach has broader applicability since it can predict PPGRs to arbitrary unseen meals, whereas the “expert”-based approach will always require CGM measurements of the meals it prescribes. 不太理解，predictor不是基于所有食物的CGM等数据学习来的？
Since our predictor also incorporates context-specific factors (e.g., previous meal content, time since sleep), this result also suggests that such factors may be important determinants of PPGRs. 我想说考虑的因素越多肯定越全面了。。。

2.9 Alterations in gut microbiota following personally tailored dietary interventions

Previous studies showed that even short-term dietary interventions of several days may significantly alter the gut microbiota. 所以说之前进行一周的饮食干预，是可以分析肠道菌群变化的。
These findings demonstrate that while both baseline microbiota composition and personalized dietary intervention vary between individuals, several consistent microbial changes may be induced by dietary intervention with a consistent effect on PPGR.