文章：Persistent gut microbiota immaturity inmalnourished Bangladeshi children

杂志：Nature   2014

研究目的：The relationships between the type of nutritional intervention, the gut microbiota, and therapeutic responses are unclear. 对于severe acute malnutrition（SAM）、moderate acute malnutrition（MAM）的干预，干预后肠道菌群的变化均不清楚。

研究方案：

       构建两个指标判断肠道菌群的状态，relative microbiota maturity index和microbiota-for-age Z-score。两个值的计算公式，relative microbiota maturity＝microbiota age of child－microbiota age of healthy children of similar chronologic age，microbiota-for-age Z score（MAZ）＝

   在两个评断指标中，有个重要变量是microbiota age，为此此文首先基于健康小孩的数据，利用机器学习方法建立年龄与肠道菌群的关系模型；

   基于两个指标观察SAM、MAM在不同阶段的肠道菌群情况。

研究结果：

  1） 年龄与肠道菌群关系模型的构建，we monitored 50 healthy Bangladeshi children monthly during the first 2 years after birth (25 singletons, 11 twin pairs, 1 set of triplets; 996 faecal samples collected monthly). By identifying bacterial taxa that discriminate the microbiota of healthy children at different chronologic ages（选择在不同年龄段存在差异的taxa）, we were able to test our hypothesis by studying 6 to 20-month-old children presenting with SAM, just before, during, and after treatment with two very different types of food intervention, as well as children with MAM.  

   训练集：in a training set of 12 children who exhibited consistently healthy anthropometric scores

   属性筛选：The relative abundances of 1222 97%-identity OTUs that passed our filtering criterion were regressed against the chronologic age of each child at the time of faecal sample collection using the Random Forests machine learning algorithm. Ranked lists of all bacterial taxa, in order of ‘age-discriminatory importance’, were determined by considering those taxa, whose relative abundance values when permuted have a larger marginal increase in means quared error. The 24 most age-discriminatory taxa identified by Random Forests are shown in Fig. 1a in rank order of their contribution to the predictive accuracy of the model and were selected as inputs to a sparse 24-taxon model.

  验证集：To test the extent to which this sparse model could be applied, we applied it, with no further parameter optimization, to additional monthly faecal samples collected from two other healthy groups of children: 13 singletons and 25 children from a birth-cohort study of twins and triplets.

   模型效果如下图

  2）干预后，SAM和MAM的肠道菌群变化（两个指数的变化），结果如下图

   可以借鉴两个指数的思路来表征肠道菌群的健康程度；

   165 of these 220 97%-identity OTUs were significantly diminished in the microbiota of children with SAM during the longer term follow-up period in both treatment groups. 这些差异的菌群未来或许可以作为营养调节的关键点。