Table 5 The impact of the newly implemented removal of unique combinations and the resulting increase in rescued features/proteins in the output matrix (without prior sorting). The old and new HarmonizR versions are shown as well as raw ComBat and limma. The protein increase is shown in absolute numbers and percentage-wise. No blocking (and therefore no sorting) has been applied during the runs shown, which achieved maximal data rescue
Dataset | Raw ComBat/limma | Proteins in the output with the original HarmonizR version (considered proteins /total proteins) | Proteins in the output with the updated HarmonizR version: removal of unique combinations (considered proteins /total proteins) | Protein increase between old and new version |
|---|---|---|---|---|
Krug et al. 2020 | 6744/13,793 (48.9%) | 10,808/13,793 (78.4%) | 13,740/13,793 (99.6%) | 2932 (27.1%) |
Petralia et al. 2020 | 3749/9155 (40.9%) | 5216/9155 (57.0%) | 9154/9155 (99.9%) | 3938 (75.5%) |
Nusinow et al. 2020 | 5098/12970 (39.3%) | 6360/12970 (49.0%) | 12,970/12,970 (100%) | 6610 (103.9%) |
Xin et al. (scRNA-seq) 2016 | 12,230/39,851 (30.7%) | 24,899/39,851 (62.5%) | 25,685/39,851 (64.5%) | 786 (3.2%) |