nucleR (version 2.4.0)

pcKeepCompDetect: Auto detection of a fitted pcKeepComp param for filterFFT function

Description

This function tries to obtain the minimum number of components needed in a FFT filter to achieve or get as close as possible to a given correlation value. Usually you don't need to call directly this function, is used in filterFFT by default.

Usage

pcKeepCompDetect(data, pc.min=0.01, pc.max=0.1, max.iter=20, verbose=FALSE, 
			cor.target=0.98, cor.tol=1e-3, smpl.num=25, smpl.min.size=2^10, smpl.max.size=2^14)

Arguments

data

Numeric vector to be filtered

pc.min, pc.max

Range of allowed values for pcKeepComp (minimum and maximum), in the range 0:1.

max.iter

Maximum number of iterations

verbose

Extra information (debug)

cor.target

Target correlation between the filtered and the original profiles. A value around 0.99 is recommeded for Next Generation Sequencing data and around 0.7 for Tiling Arrays.

cor.tol

Tolerance allowed between the obtained correlation an the target one.

smpl.num

If data is a large vector, some samples from the vector will be used instead the whole dataset. This parameters tells the number of samples to pick.

smpl.min.size, smpl.max.size

Minimum and maximum size of the samples. This is used for selection and sub-selection of ranges with meaningful values (i,e, different from 0 and NA). Power of 2 values are recommended, despite non-mandatory.

...

Parameters to be pass to autoPcKeepComp

Value

Fitted pcKeepComp value

Details

This function predicts a suitable pcKeepComp value for filterFFT function. This is the recommended amount of components (in percentage) to keep in the filterFFT function to obtain a correlation of (or near of) cor.target. The search starts from two given values pc.min, pc.max and uses linial interpolation to quickly reach a value that gives a corelation between the filtered and the original near cor.target within the specified tolerance cor.tol.

To allow a quick detection without an exhaustive search, this function uses a subset of the data by randomly sampling those regions with meaningful coverage values (i,e, different from 0 or NA) larger than smpl.min.size. If it's not possible to obtain smpl.max.size from this region (this could be due to flanking 0's, for example) at least smpl.min.size will be used to check correlation. Mean correlation between all sampled regions is used to test the performance of the pcKeepComp parameter.

If the number of meaningful bases in data is less than smpl.min.size * (smpl.num/2) all the data vector will be used instead of using sampling.

Examples

	#Load dataset
	data(nucleosome_htseq)
	data = as.vector(coverage.rpm(nucleosome_htseq)[[1]])

	#Get recommended pcKeepComp value
	pckeepcomp = pcKeepCompDetect(data, cor.target=0.99)
	print(pckeepcomp)

	#call filterFFT
	f1 = filterFFT(data, pcKeepComp=pckeepcomp)

	#Also this can be called directly
	f2 = filterFFT(data, pcKeepComp="auto", cor.target=0.99)

	#Plot
	plot(data[1:2000], col="black", type="l", lwd=2)
	lines(f1[1:2000], col="red", lwd=2)
	lines(f2[1:2000], col="blue", lwd=2, lty=2)
	legend("bottom", c("original", "two calls", "one call"), col=c("black", "red", "blue"), lty=c(1,1,2), horiz=TRUE, bty="n")