Table 1 Comparison of different peak-calling algorithms
Method | A | B | C | D | E | F | G |
|---|---|---|---|---|---|---|---|
CSPF | control or IP only | read length no orientation | merge strands no shift | N | simple height criteria | ROC curve (empirically) | both |
XSET | IP only | fragment length orientation | merge strands no shift | Y | simple height criteria | FDR estimate using Poisson distribution | both |
Mikkelsen et al. | IP only | no orientation | no merge no shift | Y | p-values from permutations | no official FDR | both |
MACS | control or IP only | fragment length orientation no duplicated reads | shift reads merge strands | N | Poisson p-values | FDR estimate by peaks in control:IP | both |
QuEST | control | orientation | shift reads merge strands | N | kernel density estimation | FDR estimate by permutations of the control | better for TF |
FindPeaks | IP only | fragment length orientation | no merge no shift | N | simple height criteria | FDR estimate by permutations of the IP | both |
SISSR | control or IP only | fragment length orientation | no merge no shift | N | compares reads on different strands | FDR estimate by peaks in background:IP | better for TF |
Kharchenko et al. | control | orientation | no merge no shift | N | Poisson distribution | FDR estimate by permutations of the control | better for TF |
PeakSeq | control | fragment length orientation | merge strands | Y | sample normalisation Binomial distribution | FDR estimate, q-values (BH correction) | both |
BayesPeak | control or IP only | fragment length orientation | no merge no shift | N | negative binomial distribution, Bayesian posterior probabilities | posterior enrichment probabilities | both |