calculateDiffMeth() shows no difference before or after pool()?

347 views
Skip to first unread message

three

unread,
Oct 14, 2016, 3:19:58 PM10/14/16
to methylkit_discussion
Hi all,

I want to test the DMR difference before and after pool(), but the results are all the same.

My data have two groups and three replicates for each group. I used the following three ways:

(1) without pool()
tileDiff=calculateDiffMeth(tiles.unit)

(2) with pool()
tiles.unit=pool(tiles.unit,sample.ids=ids)
tileDiff=calculateDiffMeth(tiles.unit,covariates=NULL,overdispersion=c("none"),adjust=c("SLIM"),effect=c("wmean"),test=c("F"))

(3) with pool() and test=c("Chisq")
tileDiff=calculateDiffMeth(tiles.unit,covariates=NULL,overdispersion=c("none"),adjust=c("SLIM"),effect=c("wmean"),test=c("Chisq"))

All the above gave results like the following with the exactly same numbers.
   chr   start     end strand     pvalue    qvalue   meth.diff
1 chr1 3001001 3002000      * 0.94478442 0.8907741   0.7246377
2 chr1 3007001 3008000      * 0.01449283 0.1666293 -13.1578947
3 chr1 3010001 3011000      * 1.00000000 0.8907741   0.0000000
4 chr1 3011001 3012000      * 1.00000000 0.8907741   0.0000000
5 chr1 3012001 3013000      * 0.39701260 0.6484307 -15.7142857
6 chr1 3013001 3014000      * 0.53387056 0.7275555   7.6998051

Is there anything wrong with my code? I was expecting at least some difference using pooled data or not.

Thanks!



Altuna Akalin

unread,
Oct 14, 2016, 5:53:56 PM10/14/16
to methylkit_...@googlegroups.com
you are probably over-writing tileDiff ?

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

three

unread,
Oct 16, 2016, 12:00:25 PM10/16/16
to methylkit_discussion
Just in case, I did that again, still the same.

#not pool
filtered.covs=filterByCoverage(covs,lo.count=9,lo.perc=NULL,hi.count=NULL,hi.perc=100)
tiles=tileMethylCounts(filtered.covs,win.size=1000,step.size=1000)
tiles.unit=unite(tiles,min.per.group=1L)
tileDiff1=calculateDiffMeth(tiles.unit)
head(tileDiff1)

>tiles.unit
methylBase object with 93564 rows
--------------
   chr   start     end strand coverage1 numCs1 numTs1 coverage2 numCs2 numTs2 coverage3 numCs3
1 chr1 3001001 3002000      *         9      7      2         9      8      1        NA     NA
2 chr1 3007001 3008000      *        20     18      2        NA     NA     NA        18     15
3 chr1 3010001 3011000      *        NA     NA     NA        NA     NA     NA        11     11
4 chr1 3011001 3012000      *         9      8      1        NA     NA     NA        NA     NA
5 chr1 3012001 3013000      *        NA     NA     NA        NA     NA     NA        10      2
6 chr1 3013001 3014000      *        27     17     10        NA     NA     NA        NA     NA
  numTs3 coverage4 numCs4 numTs4 coverage5 numCs5 numTs5 coverage6 numCs6 numTs6
1     NA         9      7      2        25     23      2        12      8      4
2      3        10     10      0         9      9      0        10     10      0
3      0        11     11      0        NA     NA     NA        NA     NA     NA
4     NA         9      9      0         9      7      2        NA     NA     NA
5      8        NA     NA     NA        14      5      9        NA     NA     NA
6     NA         9      5      4        NA     NA     NA        29     16     13
--------------

#pool
ids=c('s1','s2')
tiles.unit.p=pool(tiles.unit,sample.ids=ids)
tileDiff2=calculateDiffMeth(tiles.unit.p,covariates=NULL,overdispersion=c("none"),adjust=c("SLIM"),effect=c("wmean"),test=c("F"))
head(tileDiff2)

> tiles.unit.p
methylBase object with 93564 rows
--------------
   chr   start     end strand coverage1 numCs1 numTs1 coverage2 numCs2 numTs2
1 chr1 3001001 3002000      *        18     15      3        46     38      8
2 chr1 3007001 3008000      *        38     33      5        29     29      0
3 chr1 3010001 3011000      *        11     11      0        11     11      0
4 chr1 3011001 3012000      *         9      8      1        18     16      2
5 chr1 3012001 3013000      *        10      2      8        14      5      9
6 chr1 3013001 3014000      *        27     17     10        38     21     17
--------------

#pool and chisq.test
tileDiff3=calculateDiffMeth(tiles.unit.p,covariates=NULL,overdispersion=c("none"),adjust=c("SLIM"),effect=c("wmean"),test=c("Chisq"))
head(tileDiff2)

df1=as(tileDiff1,'data.frame')
df2=as(tileDiff2,'data.frame')
df3=as(tileDiff3,'data.frame')

write.table(df1,'df1.txt')
write.table(df2,'df2.txt')
write.table(df3,'df3.txt')

> identical(df1,df2)
[1] FALSE  *********
> identical(df1,df3)
[1] FALSE
> identical(df2,df3)
[1] TRUE

Using identical(), using test(F) or test(chisq) output the same results.
It seems that the results of 'not pool' and 'pool' are different, but the difference is really really tiny. I saved the results and checked the lines between two files.

For example, the following are one line from df1 and the corresponding line from df2. Although there is difference, but it is almost the same, the only difference is 0.154710052279306 ~ 0.154710052279307.

"57806" "chr1" 119203001 119204000 "*" 0.0123833077725416 0.154710052279306 16.6666666666667
"57806" "chr1" 119203001 119204000 "*" 0.0123833078075916 0.154710052279307 16.6666666666667

It seems really strange. I suspected that even for the not pooled data, calculateDiffMeth() automatically pool the data first. But in the manual, it said that replicates are considered. So I'm confused. Why this happened?


Another problem I encountered is about the filtering.
At first I used filterByCoverage(lo.count=9) to filter my data. This will filter each sample (each single column in the data) to remove positions with low reads. Then I thought that it may make more sense to use the total coverage from all samples to do filtering but not the coverage of one sample. And both filtering may generate different results but will not be too huge.  
For example, I have two groups and each group has three replicates. 
WAY1: I used  filterByCoverage(lo.count=9) to filter data and get DMR. 
WAY2: I also united the 6 samples first, and filtered positions with total coverage > 50, and then get DMR. 
These two ways generated dramatically different number of DMRs under the same qvalue and meth.diff cut-off.  WAY2 got almost no DMRs (from ~39021 tiles  (window=step=1000)). But WAY1 got ~2500 DMRs from ~90000 tiles. 
Is there any reason for such huge difference? I was expecting at least similar number of DMRs from WAY2.

Thank you!


在 2016年10月14日星期五 UTC+2下午11:53:56，Altuna Akalin写道：
To post to this group, send email to methylkit_...@googlegroups.com.

Altuna Akalin

unread,
Oct 16, 2016, 12:20:17 PM10/16/16
to methylkit_...@googlegroups.com
please send a reproducible example, the google groups page to this group has a link on how to produce a good reproducible example

To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.

Altuna Akalin

unread,
Oct 16, 2016, 12:23:43 PM10/16/16
to methylkit_...@googlegroups.com
To answer your second question, Way2 and Way1 are completely different ways to filter, you can't expect them to give same results

three

unread,
Oct 17, 2016, 7:12:53 AM10/17/16
to methylkit_discussion
Hi Altuna,

This is an example using the demo data in methylKit.  There is still very tiny difference in the three ways of calculateDiffMeth().

#use demo data
library('methylKit')
data(methylKit)
tiles=tileMethylCounts(object=methylRawList.obj,win.size=1000,
                                 step.size=1000,cov.bases=0)


tiles.unit=unite(tiles,min.per.group=1L)


#not pool
tileDiff1=calculateDiffMeth(tiles.unit)
head(tileDiff1)


#pool and fisher.test

ids=c('s1','s2')
tiles.unit.p=pool(tiles.unit,sample.ids=ids)
tileDiff2=calculateDiffMeth(tiles.unit.p,covariates=NULL,overdispersion=c("none"),adjust=c("SLIM"),effect=c("wmean"),test=c("F"))
head(tileDiff2)


#pool and chisq.test
tileDiff3=calculateDiffMeth(tiles.unit.p,covariates=NULL,overdispersion=c("none"),adjust=c("SLIM"),effect=c("wmean"),test=c("Chisq"))
head(tileDiff2)


df1=as(tileDiff1,'data.frame')
df2=as(tileDiff2,'data.frame')
df3=as(tileDiff3,'data.frame')



identical(df1,df2) #F
identical(df1,df3) #F
identical(df2,df3) #T (no difference between Fisher.test and chisq.test


#save to file (just very tiny difference between df1 [not pool] and df2 [pool])

write.table(df1,'df1.txt')
write.table(df2,'df2.txt')
write.table(df3,'df3.txt')






在 2016年10月16日星期日 UTC+2下午6:20:17，Altuna Akalin写道：
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

three

unread,
Oct 17, 2016, 7:21:03 AM10/17/16
to methylkit_discussion
But WAY2 gave no statisticall significant DMR from the 30,000+ tiles, this is hardly expected.
I also test different lo.count to see the number of DMRs (under the same qvalue=0.05, meth.diff=25), it seems that the number of DMRs varies a lot with different lo.count. 
Is there any practical suggestion to use what window size and lo.count? 

PS. my version is: methylKit_0.99.4

在 2016年10月16日星期日 UTC+2下午6:23:43，Altuna Akalin写道：
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

Altuna Akalin

unread,
Oct 17, 2016, 9:38:14 AM10/17/16
to methylkit_...@googlegroups.com
 In all case 2 and 3 you are pooling and this means you are applying fisher's exact test in these cases, when there is one sample per group fisher's exact test is used regardless of the other arguments you put

To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.

Altuna Akalin

unread,
Oct 17, 2016, 9:47:24 AM10/17/16
to methylkit_...@googlegroups.com

But WAY2 gave no statisticall significant DMR from the 30,000+ tiles, this is hardly expected.
I also test different lo.count to see the number of DMRs (under the same qvalue=0.05, meth.diff=25), it seems that the number of DMRs varies a lot with different lo.count. 
Is there any practical suggestion to use what window size and lo.count? 


I can suggest that you use pool() if you are using min.per.group=1L in unite() function. In this case (when min.per.group=1L ), there will be tests where you have only one replicated per group which is not how the logistic regression based tests expect. There, pooling and using fisher's exact is a better solution. 

I can't comment on the best window size but 1000bp or 500 seems reasonable judging from the studies showing CpGs that are in close vicinity behave similarly. 

lo.count we use 10, but when comparing the number of DMRs based on the coverage cutoff you should compare not absolute numbers but ratio of DMRs to tiles. 

Best,
Altuna




 
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.

three

unread,
Oct 17, 2016, 10:34:41 AM10/17/16
to methylkit_discussion
Thank you, Altuna!

But even I used min.per.group=2L or min.per.group=NULL for the demo data, the results of these three ways are still same or almost same. For example the following two lines from way1 and way2, only very slight difference (0.000174993430555084 ~ 0.000174993430555098).
"57" "chr21" 14000001 14001000 "*" 0.000218216438400963 0.000174993430555084 -23.6536536536536
"57" "chr21" 14000001 14001000 "*" 0.000218216438400963 0.000174993430555098 -23.6536536536536

I suppose that If different statistial tests are used for pooled or non-pooled data, the results (pvalue) should not be so similar. So I'm really confused about the results and don't know what parameters should I use for my data now. It seems no difference using replicate data or pooled data. Why?

About the number of DMRs, if I count the DMR#/tile# ratio, it is more stranger. I test many combinations of window size, lo.count for my own data, still could not find any clue on how DMR# varies with window size or lo.count. It is not the case that more tiles more potential DMRs. For example, the following DMR# I got with different window size. For win=500, although the tile# is the largest, the DMR# is the lowest.
----------------------------------------------------------------------
win nTile nHyper nHypo
500 104114 523 705
1000 93564 960 1421
1500 81615 1150 1662
2000 71444 1157 1761
2500 62611 1101 1666
----------------------------------------------------------------------

For different lo.count, it is also the case. I suspected that too small window size or too low lo.count may cause more candidate regions with low coverage and/or observations, then these regions will have very large pvalue. Consequently, the adjusted pvalue (qvalue) for all candidate regions will get higher, then I will get fewer significant DMR#. Is this true?  For the way to choose win and lo.count, I will test different combination of win and lo.count and pick the one with most (reasonable) number of significant DMRs.
----------------------------------------------------------------------
lo.count nTile qvalue diff nHyper nHypo
3 185839 0.05 25 18 32
5 181131 0.05 25 215 401
7 159390 0.05 25 802 1232
9 93564 0.05 25 960 1421
10 57138 0.05 25 666 978
12 14475 0.05 25 247 332
----------------------------------------------------------------------

在 2016年10月17日星期一 UTC+2下午3:47:24，Altuna Akalin写道：


To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

Altuna Akalin

unread,
Oct 17, 2016, 11:26:54 AM10/17/16
to methylkit_...@googlegroups.com
I will look into the discrepancy between statistical methods, it will take a couple of days at least. Will update here when I have figured it out

Best,
Altuna

To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.

three

unread,
Oct 17, 2016, 12:17:04 PM10/17/16
to methylkit_discussion
Thank you very much, Altuna. Looking forward to your reply!

在 2016年10月17日星期一 UTC+2下午5:26:54，Altuna Akalin写道：
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

Altuna Akalin

unread,
Oct 17, 2016, 7:17:11 PM10/17/16
to methylkit_...@googlegroups.com
On Mon, Oct 17, 2016 at 4:34 PM, three <wuth...@gmail.com> wrote:
Thank you, Altuna!

But even I used min.per.group=2L or min.per.group=NULL for the demo data, the results of these three ways are still same or almost same. For example the following two lines from way1 and way2, only very slight difference (0.000174993430555084 ~ 0.000174993430555098).
"57" "chr21" 14000001 14001000 "*" 0.000218216438400963 0.000174993430555084 -23.6536536536536
"57" "chr21" 14000001 14001000 "*" 0.000218216438400963 0.000174993430555098 -23.6536536536536


This is because in the current version fisher's exact test is not executed there will be a fix on that soon, thanks for bringing this up. And if pooled version and the original version of the methylation information goes into the logistic regression test they will get the same p-value when there is no over-dispersion correction. Fisher's exact test p-values will not be extremely different from what you get now, but they won't be identical. 
In general, I'm not sure about the point of looking at number of DMRs with different tile sizes and lo.count cutoffs. If you are trying to optimize those cutoffs, you can't do that without some sort of simulation. You need to have a setting where you have an idea about how many DMRs there should be, or you can try to optimize based on some biological knowledge. If you have differentially expressed genes maybe you can try to optimize the number of DMRs that cover the promoters of those genes.

But I will try guess why this is happening. When you are taking larger tiles, you are accumulating more counts. This means you will be able to differentiate smaller effect sizes (methylation differences) with lower p-values. You should try to use a cutoff here if you are not using. I would also look at the overlap of those regions, are they nearby ? what are p-values and differences of  500bp tiles that overlap with the significant in 1000bp tiles. If you can't pinpoint an issue that could be a bug, I will not spend more time on this.
again, you can't really optimize this without knowing correct number of DMRs beforehand. But again here, I would look if the tiles overlap. it is weird lo.count 3 and 5 give really low DMRs numbers, i would look at lo.count 10 DMRs in lo.count 3 and 5 versions. Try to see their q-values and p-values and methylation differences. Would also look at them in the browser. Also, if you are doing more tests you might pay a bigger penalty on multiple-testing. In lo.count 3, you are doing 3 times more tests than lo.count 10.

As I said, without some prior knowledge I'm not sure how one can optimize these cutoffs. 

Best,
Altuna


 
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.

Altuna Akalin

unread,
Oct 24, 2016, 2:24:36 PM10/24/16
to methylkit_...@googlegroups.com
The issue with Fisher's exact test is fixed as of version 1.1.1. It is not available in github, soon will be available in development branch of Bioconductor and I will try to fix the release version as well

three

unread,
Oct 26, 2016, 9:59:43 AM10/26/16
to methylkit_discussion
Thank you very much, Altuna. Good to know that. Sorry for the late response because I was switching to another stuff lately. 

在 2016年10月24日星期一 UTC+2下午8:24:36，Altuna Akalin写道：
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.

Altuna Akalin

unread,
Oct 26, 2016, 10:58:38 AM10/26/16
to methylkit_...@googlegroups.com
No problem. Thank you for bringing this up. Fixed version is available through GitHub and development version on bioconductor, I will try to push it to the release version as well 
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsub...@googlegroups.com.
To post to this group, send email to methylkit_discussion@googlegroups.com.

Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.


--
Sent from mobile, excuse the brevity

lian...@gmail.com

unread,
Sep 10, 2017, 11:05:40 AM9/10/17
to methylkit_discussion
A follow-up question:

In methylKit, pool () and  Correcting for overdispersion must be working together?  I also have two groups of data, each of them have three replicates (samples) from WGBS.  Unfortunately, there are big coverage differences among three samples within one of the two groups, while the other group has consistent coverage among its three samples. Can I use overdispersion correction without pooling?

In RNA-Seq analysis, normalization will take care of coverage differences among samples within the same groups. I wonder do you have a similar step in methylKit. 

Your suggestion will be highly appreciated. 

Altuna Akalin

unread,
Sep 10, 2017, 12:13:07 PM9/10/17
to methylkit_...@googlegroups.com
pooling and overdispersion correction are different things. Once you pool your samples you won't be able to use overdispersion correction. the logistic regression takes the coverage in the sense where within a group the methylation proportions that are covered more will have higher weight when fitting the model.

Best,
Altuna


To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.
To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_...@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "methylkit_discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to methylkit_discussion+unsubscrib...@googlegroups.com.

To post to this group, send email to methylkit_discussion@googlegroups.com.
Visit this group at https://groups.google.com/group/methylkit_discussion.
For more options, visit https://groups.google.com/d/optout.


--
Sent from mobile, excuse the brevity
Reply all
Reply to author
Forward
0 new messages