Figure 9

Evaluation of different normalization methods across 20 simulated metagenomic samples. Scatter plots are shown, comparing the average copy number of each KO in each sample to the corrected abundance of the KO in each sample as calculated by compositional normalization (A) and by MUSiCC (B). Each sample is represented by a different color and a regression line for each sample is illustrated. The markedly reduced variability between the slopes of the regression lines in (B) (0.99 ± 0.022; CoV = 0.022 for MUSiCC vs. 2226 ± 367; CoV = 0.165 for compositional normalization) highlights the beneficial impact of MUSiCC in reducing spurious inter-sample variation and its overall accuracy in inferring underlying copy numbers. (C) Comparison of the observed coefficient of variation (CoV) in estimating the abundance of the flagellar assembly pathway across simulated samples using different normalization methods. Notably, in this set of samples, this specific pathway should exhibit no variation. Each dot represents the CoV calculated based on a random subset of 10 samples and the box plots illustrate the distribution of CoV values across 100 such subsets. Each box plot represents the 25th and 75th percentiles, with the median in black and whiskers extending to approximately ±2.7 standard deviations. The red dot represents the CoV calculated when using all 20 simulated samples.