Colorectal cancer (CRC) is one of the most prevalent cancers in the world, especially in developed countries. In different studies, the association between CRC and dysbiosis of gut microbiome has been reported. However, most of these works focus on the taxonomic variation of the microbiome, which presents little, if any, functional insight about the reason behind and/or consequences of microbiome dysbiosis. In this study, we used a previously reported metagenome dataset which is obtained by sequencing 156 microbiome samples of healthy individuals as the control group (Co), as well as microbiome samples of patients with advanced colorectal adenoma (Ad) and colorectal carcinoma (Ca). Features of the microbiome samples have been analyzed at the level of species, as well as four functional levels, i.e., gene, KEGG orthology (KO) group, Enzyme Commission (EC) number, and reaction. It was shown that, at each of these levels, certain features exist which show significant changing trends during cancer progression. In the next step, a list of these features were extracted, which were shown to be able to predict the category of Co, Ad, and Ca samples with an accuracy of >85%. When only one group of features (species, gene, KO group, EC number, reaction) was used, KO-related features were found to be the most successful features for classifying the three categories of samples. Notably, species-related features showed the least success in sample classification. Furthermore, by applying an independent test set, we showed that these performance trends are not limited to our original dataset. We determined the most important classification features at each of the four functional levels. We propose that these features can be considered as biomarkers of CRC progression. Finally, we show that the intra-diversity of each sample at the levels of bacterial species and genes is much more than those of the KO groups, EC numbers, and reactions of that sample. Therefore, we conclude that the microbiome diversity at the species level, or gene level, is not necessarily associated with the diversity at the functional level, which again indicates the importance of KO-, EC-, and reaction-based features in metagenome analysis. The source code of proposed method is freely available from https://www.bioinformatics.org/mamed.

中文翻译：

---

超越微生物组的分类分析：重新审视结直肠癌微生物组变化的功能方法。

结直肠癌（CRC）是世界上最常见的癌症之一，尤其是在发达国家。在不同的研究中，已经报道了结直肠癌与肠道微生物群失调之间的关联。然而，这些工作大多数都集中在微生物组的分类变异上，对于微生物组生态失调背后的原因和/或后果，几乎没有提供任何功能性的见解。在本研究中，我们使用了之前报道的宏基因组数据集，该数据集是通过对作为对照组 (Co) 的 156 个健康个体的微生物组样本以及晚期结直肠腺瘤 (Ad) 和结直肠癌 (Ca) 患者的微生物组样本进行测序而获得的。 。对微生物组样品的特征进行了物种水平以及基因、KEGG直系同源（KO）组、酶委员会（EC）编号和反应四个功能水平的分析。结果表明，在每个级别上，都存在某些特征，这些特征在癌症进展过程中显示出显着的变化趋势。在下一步中，提取了这些特征的列表，这些特征被证明能够预测 Co、Ad 和 Ca 样本的类别，准确率 > 85%。当仅使用一组特征（物种、基因、KO组、EC数、反应）时，发现与KO相关的特征是对三类样本进行分类的最成功的特征。值得注意的是，与物种相关的特征在样本分类中表现得最不成功。此外，通过应用独立的测试集，我们表明这些性能趋势并不局限于我们的原始数据集。我们确定了四个功能级别中每个级别最重要的分类特征。我们建议这些特征可以被视为 CRC 进展的生物标志物。最后，我们表明每个样品在细菌种类和基因水平上的内部多样性远远高于该样品的 KO 组、EC 数和反应的多样性。因此，我们得出结论，物种水平或基因水平的微生物组多样性不一定与功能水平的多样性相关，这再次表明了宏基因组分析中基于KO、EC和反应的特征的重要性。所提出方法的源代码可以从 https://www.bioinformatics.org/mamed 免费获得。