Method | Main features | Assumptions | Pros | Cons |
---|---|---|---|---|
Strength | Linear, unbiased. Based on log2 differential measurements (e.g., log2 fold changes of measurements). | The contributions from the noisy downstream measurables sum up to zero. | Intuitive. | Noisy/biased signals can artificially decrease/increase the results. |
GPI | Based on log2 differential measurements. Down-weights weak differential measurements using false non-discovery rates. | The noisy downstream measurables have low false non-discovery rates which can be used to minimize their contributions. | Intuitive. False non-discovery rate depends on the number of experimental replicates. | False non-discovery rate depends on the number of experimental replicates. |
MASS | Linear and unbiased in absolute non-log2 scale. Dependent on absolute changes in measurements. | Absolute changes in measurements are more important than relative changes. | Intuitive. | Measurements must be directly comparable across all downstream measurables. |
EPI | Based on log2 differential measurements. Up-weights strong differential measurements without using false non-discovery rates. | The downstream measurables with higher differential values should have stronger contributions than those with lower differential values. | More robust to noisy signals than Strength. Highest sensitivity to strong differential measurements. | Less intuitive. Bootstrapping is needed for calculating Uncertainty. |