Fig. 3.

Complementarity of the different statistical tests improves DRF detection.A–C, Number of DRFs with given FDR smaller than 0.01 and calculated true FDR (from ground truth) for three representative artificial data sets. The results were assessed for different amounts of missing values and show that Miss test complements and therefore improves both DRF detection and confidence for sufficiently high replicate numbers and abundance-dependence of missing values. D, Distribution of true FDRs at a given FDR of 0.01 for 1584 artificial data sets spanning 3–10 replicates, 2–100 truly regulated proteins shifted by 1, 1.5, 2 and 5 from their original random value, and 0–100 abundance-dependence μ and missing values of 0–50%. This shows that combining the different statistical tests provide optimal numbers of true positives with high confidence (low percentage of too high true FDRs) over the range of tested data sets.