segmentation: Multi-resolution segmentation

In benja0x40/MRA.TA: Multi Resolution Analysis of Tiling Arrays

Multi-resolution segmentation

Description

Compute a multi-resolution segmentation based on locally optimal enrichment scores and internal consistency of the segmentation tree.

Usage

segmentation(Yi, name, wmax = 0, wmin = 3, gamma = 1, Tw = 0)

Arguments

Yi	vector of statistical scores produced by the enrichmentScore function.
name	prefix used to generate output file names.
wmax	maximal window size to be scanned in number of Yi values. The default is length(Yi).
wmin	minimum size of segmented regions in number of Yi values. The default is 3 consecutive Yi values.
gamma	numeric value ranging from ]0;1] and acting as a stringency factor which influences both the minimum size and enrichment score of segmented regions. The segmentation stringency gets higher when gamma value gets lower. Default value is 1.0 (no effect).
Tw	numeric value ranging from ]-Inf;0] and indicating the minimal enrichment score accepted for segmented regions. This thresholding becomes more stringent when Tw value gets lower. Default value is 0 (no threshold).

Value

segmentation returns a list object with the following attributes:

time.optimization	computation time for the optimization procedure
time.MRT.Analysis	computation time for the domain fusion procedure
n.segments	total number of locally optimal segments after optimization
n.domains	total number of domains after domain fusion
n.maxresolution	number of maximum resolution domains
n.maxscale	number of maximum scale domains
file.segments	result file for the optimization procedure
file.domains	result file for the domain fusion procedure
file.maxresolution	result file for maximum resolution domains
file.maxscale	result file for maximum scale domains
Pw.min	best statistical score (pseudo p-values from combined rank-based scores) for each scanned window size

Output files

• Common structure

  All output files are tab delimited text files including two header lines.

   line 1 = analysis parameters line 2 = column names following
  lines = data table 

• file.segments

  This file contains the optimal segment data, resulting from the the optimization procedure, and defined by columns (i, w, Piw) as follows:

   i = index of the last Yi value within the optimal segment w =
  size in number of consecutive Yi values Piw = statistical score 

• file.domains, file.maxresolution, file.maxscale

  These files contain the segmented domain data, resulting from the domain fusion procedure, and defined by columns (id, container, start, end, wmin, wmax, P, i, w) as follows:

     id        = unique identifier for each domain
     container = identifier of parent domain, 0 meaning no parent
     start     = index of the first Yi value within domain
     end       = index of the last Yi value within domain
     wmin      = minimum size of included optimal segments, in number of Yi values
     wmax      = maximum size of included optimal segments, in number of Yi values
     P         = statistical score of the locally optimal segment
     i         = index of the last Yi value within the locally optimal segment
     w         = size of the locally optimal segment in number of consecutive Yi values
  

Author(s)

Benjamin Leblanc

References

Leblanc B., Comet I., Bantignies F., and Cavalli G., Chromosome Conformation Capture on Chip (4C): data processing. Book chapter in Polycomb Group Proteins: Methods and Protocols. Lanzuolo C., Bodega B. editors, Methods in Molecular Biology (2016). http://dx.doi.org/10.1007/978-1-4939-6380-5_21

See Also

calc.Qi, enrichmentScore, domainogram, plotOptimalSegments, plotDomains

Examples

# Simulate enrichment signal
n <- 2000
Mi <- rep(0, n)
Mi <- Mi + dnorm(1:n, 2.5*n/20,  n/40) + dnorm(1:n, 4*n/20,  50)
Mi <- Mi + 4 * dnorm(1:n, 5*n/10, n/10)
Mi <- Mi + dnorm(1:n, 16*n/20,  n/40) + dnorm(1:n, 17.5*n/20,  50)
Mi <- (Mi/max(Mi))^4 + rnorm(n)/4

# Compute enrichment scores
Yi <- enrichmentScore(Mi)

# Multi-resolution segmentation
seg.c <- segmentation(Yi, name="MRA_demo", wmin=20)

# Load segmentation results
opts <- read.delim(seg.c$file.segments, stringsAsFactors=F, skip=1)
doms <- read.delim(seg.c$file.domains, stringsAsFactors=F)
doms.mr <- read.delim(seg.c$file.maxresolution, stringsAsFactors=F)
doms.ms <- read.delim(seg.c$file.maxscale, stringsAsFactors=F)

# Visualization coordinates
x.s <- 1:n - 0.5
x.e <- 1:n + 0.5
w2y <- wSize2yAxis(n, logscale=T)

layout(matrix(1:2, 2, 1), heights=c(3,1)/4)
par(mar=c(3, 4, 1, 2)) # default bottom, left, top, right = c(5, 4, 4, 2)

# Plot domainogram
domainogram(Yi, x.s, x.e, w2y)
plot(Mi, type='l', xaxs = 'i')

# Visualize segmentation results
plotOptimalSegments(opts, x.s, x.e, w2y, col="black")
plot(Mi, type='l', xaxs = 'i')

# Visualize multi-resolution domains
plotDomains(doms, x.s, x.e, w2y, col=rgb(0,0,0,0.5), border=rgb(0,0,0,0))
# Visualize max. resolution and max. scale domains
plotDomains(doms.mr, x.s, x.e, w2y, col=rgb(0,1,0,0.5), border=rgb(0,0,0,0), lwd=2, lty=1, add=T)
plotDomains(doms.ms, x.s, x.e, w2y, col=rgb(1,0,0,0.5), border=rgb(0,0,0,0), lwd=2, lty=1, add=T)
legend("topright", c("Max. resolution", "Max. scale", "Both"), fill=c("green", "red", "chocolate"), bty='n')
plot(Mi, type='l', xaxs = 'i')

benja0x40/MRA.TA documentation built on March 13, 2023, 5:15 a.m.