Fig. 4

Relative error on Richness (A) and Pielou (B) alpha diversity indices for the three simulated datasets. (The lower the better). Relative errors are computed between alpha indices value of ground truth data and preprocessed data are shown. Results are aggregated according to the imputation method used in the pipelines. “None” identifies pipelines where no imputation is performed (i.e. raw and normalized-only data). Distributions of relative error values that result statistically lower (one-sided paired Mann–Whitney paired U-test, Benjamini–Hochberg correction, significant threshold 0.05) than relative errors calculated on raw/normalized-only data are indicated with the symbol “*”, followed by the interpretation of Cohen’s d effect size (N: negligible, VS: very small, S: small, M: medium, L: large, VL: very large, H: huge). The vertical dashed line indicates the median relative error on raw/normalized-only data