k- means clustering of chipseq peak data wrt to TSS

[honey]

This question has been discussed number of times in this forum- I have a paculiar situation. I have chips seq data for normal and cancer cells I have used a tool to call peaks and now has bed files which shsiblow enrichment in normal as well as cancer cells. I want to perform K means clustering using TSS as refrence of these peaks. I have used Seqminer to perform clustering I do have results which specifically show that different clusters for normal and cancer cells. The problem is I cannot get a good quality picture. I have tried all differenrt (everything) possible combinations. but picyure come out be very fuzzy. I used 5KB+/- t0 500bp+/- various other adjustments, of no use.

I want to generate fig5 of http://nar.oxfordjournals.org/conten...ar.gkr416.full
or Fig 2E of http://www.ncbi.nlm.nih.gov/pubmed/18992931
Fig1A of http://www.jbc.org/content/early/201...bc.M111.266254
I am also open to use any other tool which can allow me to use bed files for doing K means clustering. I want to stick with my peak calling alogrithm. I undertsand MACS in Galaxy can give such clustering but I have to use MACs peak calling to use such that tool.

Thanks for your help.

[kadircaner]

Try heatmap module under cistrome.org/ap. You can provide your own bed file along with wig files that you want to display. Picture quality is better than seqminer...

[honey]

k- means clustering of chipseq peak data wrt to TSS

Part of the problem is I dont have wiggle file and has only bed file with peak score. Moreover, Cistrome will not allow me to get list of candidate peaks in each cluster.

[frozenlyse]

This can be done using Repitools from bioconductor - see section 5.3 of the vignette for an example.

[honey]

K means on Chipseq

frozenlyse,
Thanks

Let me clarify what I have I have two bed files with following structure following the Vignette.
Chr start stop strand Reads Per Million (RPM) Transcript ID Gene Symbol normal: reads _cancer: reads
chr4 56953396 56946396 - 5.04223 NM_181806 AASDH 35 106
chr4 462999 455999 - 4.20186 NR_002451 ABCA11P 26 96
chr7 150560251 150553251 - 4.97758 NM_007189 ABCF2 43 115
chr21 42513151 42589572 + 3.55542 NM_207174 ABCG1 4 57
chr4 8216460 8209460 - 3.6847 NM_001130088 ABLIM2 12 51
chr3 130076023 130083023 + 4.46043 NR_033426 ACAD9 34 96
chr3 133518991 133569357 + 4.2665 NM_001134194 ACPP 44 123

One is for cancer and other is for normal. The coordinates are different for cancer and normal cells (above is snapshot of cancer bed file only). The question is if I have to do K means to show that normal and cancer show different type of clusters. Should I be taking the normal and cancer in one data frame (like Vignette). Then clustering will be based on raw reads(which perhaps may not be ideal) or RPM ( I get only for one sample- normal or cancer; further it may be one gene may be enriched in cancer but not in normal and vice versa). I looked for data(Hi1amples) from 4.3 but could not fine find structure of H1samples.

I understand it appear that I am asking something which has to readymade, but it is kind of frustrating that I have results in seqminer but no good fig and I have not got any feedback from there.

Thanks

[honey]

error in running Repitools

I tried to run the package as mentioned in fig5 of http://nar.oxfordjournals.org/conten...ar.gkr416.full
but is getting error-



I have changed upstream downstream distance, frequency as well as smoothing and no smoothing.
Thanks for your help

Error in pam(all, n.clusters, cluster.only = TRUE, do.swap = FALSE) :
NA/NaN/Inf in foreign function call (arg 4)
In addition: Warning messages:
1: In min(x) : no non-missing arguments to min; returning Inf
2: In max(x) : no non-missing arguments to max; returning -Inf