【摘要】 宏基因组组装往往只能得到较长片段的叠连群，无法恢复完整的基因组。现有的一些分箱方法并未充分挖掘叠连群序列组成和样本覆盖度内部结构信息。开发了基于流形嵌入的宏基因组学叠连群分箱方法，可以挖掘出高维数据中内部的非线性结构特征，从而降低数据的维度，提高计算性能。使用流形嵌入的结果估计出初始分箱数，比使用基于单拷贝基因的分箱数初始化方法更为高效。基于序列组成和样本覆盖度信息，流形嵌入更好地表现出了高维数据嵌入空间的内部结构，为分箱器提供了更有效的特征信息。实验对比了其他方法，结果表明所提方法在SpeciesMock数据集上达到了最高的准确率(ACC)、归一化互信息(NMI)和归一化兰德指数(ARI)。更多还原

【Abstract】 Metagenomics assembling can only obtain long segments of contigs, and cannot restore the complete genomes. Some existing binning methods do not fully mine the internal structure information of sequence composition and sample coverage of contigs. A metagenomics contig binning method based on manifold embedding is developed, which can mine the internal nonlinear structural features in high-dimensional data, so as to reduce the dimension of data and improve computational performance. It used the results of manifold embedding to estimate the initial bin number, which was more efficient than the bin number initialization method based on single copy genes. Based on the sequence composition and sample coverage information, manifold embedding better showed the internal structure of high-dimensional data embedding space, and provided more effective feature information for binning. Compared with other methods, this method achieves the highest ACC, NMI and Ari on the SpeciesMock data set.更多还原