Parameter Guidance

Most parameters can be left at their default values. The following sections provide guidance on key arguments that may be useful to adjust for specific applications.

Peak Calling Controls

matchingParams.peakMode

Specifies calling of narrow and/or broad marks.

• narrow

  Writes UCSC narrowPeak calls.

• broad

  Writes UCSC gappedPeak calls.

• both

  Writes narrow and broad calls.

matchingParams.selectionPenaltyMultiplier

• Scales the data-derived per-bin selection penalty. The default is 1.0.

• Values below 1.0 yield more relaxed selection. Values above 1.0 increase stringency.

matchingParams.gamma

• Sets the penalty for boundaries between selected and unselected bins (default 0.25). Sometimes called the total variation penalty in signal and image processing literature.

• Larger values discourage fragmentation of selected regions, while smaller values allow more granular calls.

• For instance, if calling TF ChIP-seq peaks for a punctate mark like Pol II, a matchingParams.gamma: 0.05 may be appropriate.

matchingParams.minMeanSignal

• Keeps candidate peaks whose mean signal reaches this threshold.

• The default is 0.01. Disable with null.

(Advanced) matchingParams.thresholdZ

• This parameter controls which regions are used to estimate the selection/budget penalty. Higher values restrict the estimation to more strongly enriched genomic regions. Note that this quantity need not change monotonically with respect to the final selection/budget penalty.

• Lower values allow less enriched regions to influence budget/penalty calibration. The default is 2.0

Estimation Controls

countingParams.intervalSizeBP

• Sets the genomic bin size in base pairs. Read and fragment inputs default to 50.

• Higher-resolution results may be obtained for deeply sequenced data using smaller intervalSizeBP: 25 bp, 10 bp, etc. For detecting domain-level enriched-regions in broad marks like H3K27me3, larger values (100, 250, etc.) can be used. Larger intervalSizeBP is encouraged for shallow sequencing data.

observationParams.precisionMultiplierMin and observationParams.precisionMultiplierMax

• Bound observation precision multipliers. The defaults are 0.1 and 10.0.

processParams.precisionMultiplierMin and processParams.precisionMultiplierMax

• Bound process precision multipliers. The defaults are 1.0e-4 and 10.0.

• Because \(Q_i=Q_0/\kappa_i\), the multipliers let the local frequency response adapt: smaller \(\kappa_i\) admits faster variation, whereas larger \(\kappa_i\) favors smoother variation.

outputParams.stateShrinkageEnabled

• Enables an experimental posterior state shrinkage using a spike-and-Student-\(t\) prior, with consenrich-estimated positional signal and uncertainty plugged-in.

• Disable it in the YAML config w/ outputParams.stateShrinkageEnabled: false.

• Depending on the downstream application, the shrunk state/uncertainty tracks should be interpreted as approximate since the inputs are (fixed) estimates themselves, i.e., uncertainty is not propagated in the fully Bayesian sense.

• Specifically, cutoffs on the shrunkState tracks (e.g., matchingParams.minMeanSignal for peaks) may become conservative relative to the original state.