Karimi Elham, Geslain Enora, Belcour Arnaud, Frioux Clémence, Aïte Méziane, Siegel Anne, Corre Erwan, Dittami Simon M
UMR8227, Integrative Biology of Marine Models, Sorbonne Université/CNRS, Station Biologique de Roscoff, Roscoff, France.
FR2424, Sorbonne Université/CNRS, Station Biologique de Roscoff, Roscoff, France.
PeerJ. 2021 May 6;9:e11344. doi: 10.7717/peerj.11344. eCollection 2021.
Animals, plants, and algae rely on symbiotic microorganisms for their development and functioning. Genome sequencing and genomic analyses of these microorganisms provide opportunities to construct metabolic networks and to analyze the metabolism of the symbiotic communities they constitute. Genome-scale metabolic network reconstructions rest on information gained from genome annotation. As there are multiple annotation pipelines available, the question arises to what extent differences in annotation pipelines impact outcomes of these analyses. Here, we compare five commonly used pipelines (Prokka, MaGe, IMG, DFAST, RAST) from predicted annotation features (coding sequences, Enzyme Commission numbers, hypothetical proteins) to the metabolic network-based analysis of symbiotic communities (biochemical reactions, producible compounds, and selection of minimal complementary bacterial communities). While Prokka and IMG produced the most extensive networks, RAST and DFAST networks produced the fewest false positives and the most connected networks with the fewest dead-end metabolites. Our results underline differences between the outputs of the tested pipelines at all examined levels, with small differences in the draft metabolic networks resulting in the selection of different microbial consortia to expand the metabolic capabilities of the algal host. However, the consortia generated yielded similar predicted producible compounds and could therefore be considered functionally interchangeable. This contrast between selected communities and community functions depending on the annotation pipeline needs to be taken into consideration when interpreting the results of metabolic complementarity analyses. In the future, experimental validation of bioinformatic predictions will likely be crucial to both evaluate and refine the pipelines and needs to be coupled with increased efforts to expand and improve annotations in reference databases.
动物、植物和藻类的发育与功能依赖于共生微生物。对这些微生物进行基因组测序和基因组分析，为构建代谢网络以及分析它们所构成的共生群落的代谢提供了机会。基因组规模的代谢网络重建基于从基因组注释中获得的信息。由于有多种注释流程可供使用，因此出现了一个问题，即注释流程的差异在多大程度上会影响这些分析的结果。在这里，我们比较了五种常用的流程（Prokka、MaGe、IMG、DFAST、RAST），从预测的注释特征（编码序列、酶委员会编号、假设蛋白）到基于代谢网络的共生群落分析（生化反应、可产生的化合物以及最小互补细菌群落的选择）。虽然Prokka和IMG生成的网络最为广泛，但RAST和DFAST网络产生的假阳性最少，且连接性最强，死端代谢物最少。我们的结果强调了在所有检测水平上，测试流程输出之间的差异，代谢网络草图中的微小差异会导致选择不同的微生物群落来扩展藻类宿主的代谢能力。然而，生成的群落产生了相似的预测可产生化合物，因此可以认为在功能上是可互换的。在解释代谢互补性分析结果时，需要考虑到所选群落与取决于注释流程的群落功能之间的这种差异。未来，对生物信息学预测进行实验验证可能对于评估和完善流程至关重要，并且需要加大努力来扩展和改进参考数据库中的注释。
