Abstract：
Characterization of microbial community metabolic output is crucial to understanding their functions. Construction of genome-scale metabolic models from metagenome-assembled genomes (MAG) has enabled prediction of metabolite production by microbial communities, yet little is known about their accuracy. Here, we examined the performance of two approaches for metabolite prediction from metagenomes, one that is MAG-guided and another that is taxonomic reference-guided. We applied both on shotgun metagenomics data from human and environmental samples, and validated findings in the human samples using untargeted metabolomics. We found that in human samples, where taxonomic profiling is optimized and reference genomes are readily available, when number of input taxa was normalized, the reference-guided approach predicted more metabolites than the MAG-guided approach. The two approaches showed significant overlap but each identified metabolites not predicted in the other. Pathway enrichment analyses identified significant differences in inferences derived from data based on the approach, highlighting the need for caution in interpretation. In environmental samples, when the number of input taxa was normalized, the reference-guided approach predicted more metabolites than the MAG-guided approach for total metabolites in both sample types and non-redundant metabolites in seawater samples. Nonetheless, as was observed for the human samples, the approaches overlapped substantially but also predicted metabolites not observed in the other. Our findings report on utility of a complementary input to genome-scale metabolic model construction that is less computationally intensive forgoing MAG assembly and refinement, and that can be applied on shallow shotgun sequencing where MAGs cannot be generated.IMPORTANCELittle is known about the accuracy of genome-scale metabolic models (GEMs) of microbial communities despite their influence on inferring community metabolic outputs and culture conditions. The performance of GEMs for metabolite prediction from metagenomes was assessed by applying two approaches on shotgun metagenomics data from human and environmental samples, and validating findings in the human samples using untargeted metabolomics. The performance of the approach was found to be dependent on sample type, but collectively, the reference-guided approach predicted more metabolites than the MAG-guided approach. Despite the differences, the predictions from the approaches overlapped substantially but each identified metabolites not predicted in the other. We found significant differences in biological inferences based on the approach, with some examples of uniquely enriched pathways in one group being invalidated when using the alternative approach, highlighting the need for caution in interpretation of GEMs.
摘要：
微生物群落代谢输出的表征对于理解其功能至关重要。从宏基因组组装的基因组（MAG）构建基因组规模的代谢模型已经能够预测微生物群落的代谢物产生，然而，人们对它们的准确性知之甚少。这里,我们检查了从宏基因组中预测代谢物的两种方法的性能，一个是MAG指导的，另一个是分类学参考指导的。我们应用了来自人类和环境样本的鸟枪宏基因组学数据，并使用非靶向代谢组学在人类样本中验证了发现。我们发现在人体样本中，在分类学分析优化和参考基因组是现成的，当输入类群的数量被归一化时，参考指导方法比MAG指导方法预测更多的代谢物。这两种方法显示出明显的重叠，但每种方法鉴定的代谢物在另一种方法中没有预测。路径富集分析确定了基于该方法的数据得出的推论存在显着差异，强调在解释中需要谨慎。在环境样本中，当输入类群的数量被归一化时，对于两种样品类型中的总代谢物和海水样品中的非冗余代谢物,参考指导方法比MAG指导方法预测更多的代谢物.尽管如此，正如在人类样本中观察到的那样，这些方法基本上重叠，但也预测了在其他方法中未观察到的代谢物。我们的发现报告了互补输入对基因组规模代谢模型构建的实用性，该模型在MAG组装和细化之前计算强度较低，并且可以应用于无法生成MAG的浅层鸟枪测序。IMPORTANCELittle知道微生物群落的基因组尺度代谢模型（GEMs）的准确性，尽管它们对推断群落代谢输出和培养条件有影响。通过对来自人类和环境样品的shot弹枪宏基因组学数据应用两种方法，评估了GEM用于宏基因组代谢物预测的性能。并使用非靶向代谢组学验证人类样本中的发现。发现该方法的性能取决于样本类型，但总的来说，参考指导方法比MAG指导方法预测更多的代谢物。尽管存在差异,这些方法的预测基本重叠，但每种确定的代谢物在另一种中没有预测。我们发现基于该方法的生物学推论存在显着差异，当使用替代方法时，一组中的一些独特富集途径无效的例子，强调在解释GEM时需要谨慎。
