Fig. 2: Performance of transformation methods in estimating observed richness. | Nature Communications

Fig. 2: Performance of transformation methods in estimating observed richness.

From: Benchmarking microbiome transformations favors experimental quantitative approaches to address compositionality and sampling depth biases

Fig. 2

a Correlation between the observed richness in the original synthetic communities and (transformed) sequence matrices. Correlations are classified in five categories (two-sided Pearson correlation tests): significant, strong (Pearson R > 0.8, p-value < 0.05); significant, moderate (0.8 ≥ R > 0.5 p-value < 0.05); significant, mild (0.5 ≥ R > 0, p-value < 0.05); non-significant (p-value ≥ 0.05); and negative (R < 0, p-value < 0.05). Stacked barplots represent the percentage of correlations falling into each of the categories for each method and scenario (for n = 10 simulated matrices per scenario). Method performance comparison (Pearson R median test): \(*\), two-sided Kruskal–Wallis p-value < 0.05 (Supplementary Data 2). b Pairwise evaluation of method performance based on distribution of correlation coefficients determined in panel a (Pearson R Dunn test, two-sided). Significant (p-value < 0.05) comparisons after multiple-testing correction are colored, with colors representing the sign of the effect size (Z-statistic) of each pairwise comparison. A red (blue) color indicates that the method in the corresponding row has a higher (lower) performance than the method in the corresponding column. Seq: sequencing data, RMP: relative microbiome profiling, CSS: cumulative sum scaling, GMPR: geometric mean of pairwise ratios, UQ: upper quartile, RLE: relative log expression, TMM: trimmed mean of M-values, ACS: absolute count scaling, QMP: quantitative microbiome profiling.

Back to article page