Method to find the cutoff at which gains and losses are considered significant using permutations

1 | ```
findSigLevelFdr(data, observedSpm, n = 1, fdrTarget=0.05, maxmem=1000)
``` |

`data` |
aCGH data in the same format as used for 'calcSpm' |

`observedSpm` |
A sample point matrix as produced by 'calcSpm' |

`n` |
Number of permutations |

`fdrTarget` |
Target False Discovery Rate (FDR) |

`maxmem` |
This parameter controls memory usage, set to lower value to lower memory consumption |

The number of permutations needed for reliable results depends on the data and can not be determined beforehand. As a general rule-of-thumb around 100 permutations should be used for 'quick checks' and around 2000 permutations for more rigorous testing. The FDR method is less conservatie than the p-value based approach since instead of controlling the family wise error rate (FWER, P(false positive > 1)) it controls the false discovery rate (FDR) (false positives / total number of called data points).

A list with the cutoffs corresponding to the given FDR

`pos ` |
The cutoff for the gains |

`neg ` |
The cutoff for the losses' |

Jorma de Ronde

`plotScaleSpace`

1 2 3 4 5 6 7 8 9 | ```
data(hsSampleData)
data(hsMirrorLocs)
spm1mb <- calcSpm(hsSampleData, hsMirrorLocs)
sigLevel1mb <- findSigLevelTrad(hsSampleData, spm1mb, n=3)
plot(spm1mb, sigLevels=sigLevel1mb)
plotScaleSpace(list(spm1mb), list(sigLevel1mb), type='g')
``` |

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.