using fecal microbiota composition as a non-invasive diagnostic tool to classify colorectal carcinoma and adenoma.

We profiled the microbiota of 1002 fecal samples of colorectal carcinoma and adenoma patients and normal controls from southeastern China using 16S rRNA Illumina PE sequencing. We built random forest tree based bi-class classifier of CRC and normal and multi-class classifier of cancer, adenomas, and normal, and then test them independently. Batch effects were studied and spike-in methods were proposed as an effective strategy to model such effects and improve the results of multi-class classification. Effects of confounding factors were also studied.本研究针对中国东南部地区结直肠癌（colorectal carcinoma）、腺瘤患者及正常对照人群的1002份粪便样本，采用16S rRNA（16S核糖体RNA）Illumina双端测序技术开展微生物组谱分析。本研究构建了基于随机森林树的结直肠癌与正常对照二分类分类器，以及针对结直肠癌、腺瘤与正常对照的多分类分类器，并对上述分类器开展独立性能验证。本研究同时对测序批次效应进行了分析，提出采用spike-in法作为建模此类效应、优化多分类任务分类效果的有效策略。此外，本研究还探究了各类混杂因素对实验结果的影响。