PeakSegJointFaster {PeakSegJoint}

PeakSegJointFaster

Description

Run the PeakSegJointFaster heuristic optimization algorithm, for several bin.factor parameter values, keeping only the most likely model found. This gives an approximate solution to a multi-sample Poisson maximum likelihood segmentation problem. Given S samples, this function computes a sequence of S+1 PeakSegJoint models, with 0, ..., S samples with an overlapping peak (maximum of one peak per sample). It also computes for G groups, the seq of G+1 models, with 0, ..., G groups with an overlapping peak.

Usage

PeakSegJointFaster(profiles, 
    bin.factor.vec = 2:7)

Arguments

profiles

data.frame with columns sample.id, sample.group, chromStart, chromEnd, count.

bin.factor.vec

Size of bin pyramid. Bigger values result in slower computation.

Value

List of model fit results.

Author(s)

Toby Dylan Hocking

Examples

library(PeakSegJoint)
data(H3K36me3.TDH.other.chunk1, envir=environment())
some.counts <- subset(
  H3K36me3.TDH.other.chunk1$counts,
  43000000 < chromEnd &
  chromStart < 43200000)
some.counts$sample.group <- some.counts$cell.type

fit <- PeakSegJointFaster(some.counts, 2:7)

if(interactive() && require(ggplot2)){

  both <- with(fit, rbind(
    data.frame(model="sample", sample.modelSelection),
  data.frame(model="group", group.modelSelection)))
  ggplot()+
    ggtitle("model selection functions")+
    scale_size_manual(values=c(sample=2, group=1))+
    geom_segment(aes(min.log.lambda, complexity,
                     color=model, size=model,
                     xend=max.log.lambda, yend=complexity),
                 data=both)+
    xlab("log(penalty)")+
    ylab("model complexity (samples or groups with a common peak)")

}