Description Usage Arguments Details Value Author(s) References See Also Examples
Learn the peak profile from ChIP-chip data, to be used for data deconvolution via 'chip.deconv' or 'deconv.entire.genome'.
1 2 3 4 5 6 7 8 | fit.peak.profile(data, tile.size, n.peaks = 30, n.skip = 10, in.kernel =
NA, fits = NULL, method = "Nelder-Mead", positions=c( 0, 25, 50, 100,
150, seq( 200, (mini.window*1.5)+1, by=100 ) ), re.fit = 25, start.pars
= c(shape = 7, scale = 50, bs.size = 20, h.cutoff = 15), to.be.fit=c(
"shape", "scale", "bs.size", "h.cutoff" ), rnd = F, mini.window = max(5
* tile.size, 1300), plot = T, name = "", no.multicore=T, ...)
plot.fit.peak.profile(x, n.peak.plot = 7, plot.spline = F, ...)
|
data |
Input data matrix, connection, or file name. See 'chip.deconv' for details. |
tile.size |
Probe length (bp) for use in kernel generation. See 'generate.binding.profile' for details. |
n.peaks |
Number of (biggest) peaks to learn from |
n.skip |
Skip this many worst-fitting of the 'n.peaks'. Enables filtering out of peaks that don't agree with the majority of isolated peaks. |
in.kernel |
Input seed kernel to start with (guessed if 'NA'). |
fits |
Input preliminary fits (via 'deconv.entire.genome'; generated from input data if 'NULL'). |
method |
Optimization method, see 'optim' |
positions |
Passed directly to 'generate.binding.profile' |
re.fit |
Re-run 'deconv.entire.genome' on input data using current best-fit profile, every 're.fit' iterations |
start.pars |
Starting parameters for model profile. See Details. |
to.be.fit |
Names of parameters to be learned. See Details. |
rnd |
Add some 'jitter' to the starting parameters |
plot |
Plot the fit as it progresses (including data and fit to several bright peaks in the data) |
mini.window |
Size of window to be plotted around each of the bright peaks, if 'plot' is TRUE |
no.multicore |
Prevent use of multiple cores, even if 'multicore' is installed. |
name |
Ignored |
... |
Further parameters for 'deconv.entire.genome' and 'generate.binding.profile' |
x |
Object output from 'fit.peak.profile' |
n.peak.plot |
Number of bright peaks (and their fits) to include in the plot |
plot.spline |
Plot a 'smooth.spline' fit to the data for reference |
'fit.peak.profile' iteratively runs 'deconv.entire.genome' on a data set, isolates the 'n.peaks' brightest peaks, and fits parameters to the model profile use by 'generate.binding.profile', then re-fits the data using 'deconv.entire.genome'. Currently, the parameters that are fit include the shape and scale of the Gamma function used for the fragment length distribution, the binding site footprint, and the "cutoff" for hybridization. All parameters listed in 'to.be.fit' will be optimized. Parameters not listed in 'start.pars' will be set to the defaults.
A list of class 'fit.peak.profile, for which a 'plot' function exists, and containing the following elements:
score |
The log of the RSS of the final peak profile to the 'n.peaks - 'n.skip' brightest isolated peaks in the data (this is the measure that was optimized). |
kernel |
The final peak profile, generated by 'generate.binding.profile'. |
new.fits |
The deconvolution fits of the final peak profile to the 'n.peaks' brightest peaks in the data, generated by 'chip.deconv'. |
is.bad |
Logical vector telling which of the 30 'n.peaks' brightest peaks were not among the 'n.skip' poorly-fitting peaks. |
par |
Final best-fit model parameters for the peak profile. |
peaks |
Two-column matrix listing the centers and intensities of the 'n.peaks' brightest peaks in the data, that were used for the fitting. |
args |
All parameters input to 'fit.peak.profile', for future reference. |
David J Reiss, Institute for Systems Biology
Maintainer: <dreiss@systemsbiology.org>
Reiss, DJ and Facciotti, MT and Baliga, NS. (2007). "Model-based
deconvolution of genome-wide DNA binding",
Bioinformatics; doi: 10.1093/bioinformatics/btm592.
http://baliga.systemsbiology.net/medichi
chip.deconv, deconv.entire.genome, generate.fake.data,
generate.binding.profile, MeDiChI-data,
lars
, quadprog
, Matrix
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 | ## Fit the peak profile to the high-resolution Nimblegen data,
## plotting progress. Note this will take some time.
data( "halo.hires", package="MeDiChI" )
## Not run:
params <- fit.peak.profile( data.halo.hires, tile.size=50,
quant.cutoff="q0.99", chrom="Chr",
fit.res=30, max.steps=100, plot=TRUE )
plot( params )
## Use the output kernel for deconvolution (see 'deconv.entire.genome'):
fits <- deconv.entire.genome( data.halo.lowres, fit.res=10,
n.boot=1, kernel=params\$kernel, verbose=TRUE, trace=FALSE )
## End(Not run)
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.