Hourglass distribution

August 6, 2013 1 comment

Some days ago, I went through an article on mutation rate mentioning the “hourglass distribution”. The illustration of this distribution is pretty obvious from the plot in the article, with for some chromosome a pretty clear reduction in number of SNP around centromeres. I just wondered if we could also observe this phenomenon on bos taurus. So i just had a look on dbSNP138 data to have some clues.

First version

#Retrieve dbSNP
 wget ftp://ftp.ncbi.nih.gov/snp/organisms/cow_9913/VCF/vcf_*vcf.gz

And then just run the following R code.

#Open an output file
#split the screen in 30
#Create a vector of chromosomes
 #run a loop on chromosomes
 for( i in chr){
 #Read file
 #plot the snp density
 plot(density(Pos$V2),col="grey",main=paste("BTA : ",i,sep=""))

Notes on code :

  • read.table is used directly on a gzipped file (very handy trick)
  • dbsnp file have a 15 lines long header, so I use the skip =15 option
  • I had some glitches while reading some files, (a problem with # of fields) option fill=TRUE, just fix it
  • Plot are nice….but polygon are even better, so I first plot the density and then add a polygon on it
  • rgb function is a simple way to obtain transparency, so after the three values for red, green and blue, I add the alpha setting to a value of 0.04

And after a while….you ‘ll obtain something like this


I must say I was a bit disappointed. At least, there are no clear pattern as can be seen in human. All the bovine chromosomes are acrocentric  this may explain why generally no clear decrease in SNP density can be seen. The pattern observed on chromosome 12, 18 and 10 were even more surprising. I am wondering if there could be some sampling bias. Concerning the pattern on BTA23, the latter could be due to MHC, known to exhibit a great diversity. Density computation may also blur a bit things.

Second version

The basic work being done, we can try to investigate others hypotheses. As instance, are SNP and Indel distributed the same way along the genome ? With some slight changes, the code become :

#Open an output file
#split the screen in 30
#Create a vector of chromosomes
 #run a loop on chromosomes
 for( i in chr){
 #Read file
 #plot the snp density
 plot(density(Pos$V2[grep("snp",Pos$V8,perl=TRUE)]),col="grey",main=paste("BTA : ",i,sep=""))
 #Add in-del line

Notes on code :

  • in dbsnp variant are coded either as snp or in-del, we extract line with the grep function accordingly
  • I tweaked a bit the indel line in order to avoid scale problems.

SNPandindelDistributionWe observe roughly the same pattern between snp and indel, albeit indel distribution may be smoother. I was expecting some discrepancies (relying on the article by Montgomery et al. but here again, we are only dealing with 1 base indel, which is not really representative of short indel  in general). I may try to check this results with my own results.


Recent articles

August 1, 2013 Leave a comment

Just a small review of the articles I went through recently :

  • Why proper definition of any statistic matter. An article on the numerous Fst definitions, and their respective behavior with varying assumptions : MAF spectrum, samples size etc.  The article is very comprehensive and give a nice review of main concept around Fst.
  • Still around population genetics an article of my PhD supervisor on selective sweeps identification. The supporting information are also pretty worthy.
  • A nice review on measure of dependence it’s a  real relief to see alternative to the correlation coefficient do exist ! I must admit that pointless statements like “X and Y are correlated at 5%” just kill me.
  • “Encode returns” a reply to Graur’s article by Mattick and Dinger.
  • While at first glance its title may sound rather tautological, this article on long intergenic non coding RNA nicely shows that similar behavior (Ribosome occupancy) does not imply similar results (Protein coding). I wondered if the absence of stop codon in lincRNA (and thus the absence of ribosome release) impact the cell, but apparently lincRNA are not numerous enough to have any substantial effect.

Software update

July 25, 2013 Leave a comment

So what are the last noticeable update of this latest months ?

GATK 2.6-5

At last HaplotypeCaller is now a real alternative for Big project ! <grumpy> The bad thing <\grumpy> is that now GATK need java7, which is still not so common nowadays and (thanks to oracle) no that easy to install !

I am actually running it on a batch of 100 bulls  @12&+X(so far so good, I will complete a whole genome variant calling in 2 weeks).

Beagle 4
Now allowing imputation ! And this just make everything so simple !


This tool has been released in its 5.0 version.


This is the yearly release. Not a lot of new things but it’s always worthy to upgrade.

Categories: emacs, Linux, NGS

Serial snapshots with IGV batch

July 16, 2013 Leave a comment

IGV is a very handy tool. Nevertheless, scrolling from one position to another may be fastidious. Second, the bad aspect of this very user-friendly software, is that you can spend hours looking here and there, with at last no backtrack of your discoveries.

Thankfully, IGV can run in batch mode, allowing, for a targeted list of positions, to take screenshots and store the later in a folder. We’ll illustrate in the following with a small example :

Setting up a session

To test our script, we will first download some publicly available data.

#Download a vcf file (on beagle4 website, you may have to change file name according to last release)
wget http://faculty.washington.edu/browning/beagle/test.r1099.vcf.gz
gunzip test.r1099.vcf.gz
#Create an index
igvtools index test.r1099.vcf
#(You can alternatively prefer to use igvtools from igv GUI)
#Launch igv

To check if everything is right, load the vcf file test.r1099.vcf. Then move to positions chr22:20,000,000-20,100,000.


Set everything to your taste, and save the session :  File, save session, you should obtain a xml file.

Running IGV in batch mode

Now, let’s consider you are interested in some particular position. Let’s say we’ve stored several positions of interest in a csv file. Our aim is to create an IGV batch file.

Basically, we’ll have to load the session, set a directory to store the screenshots, and then move from one position to the other. A very crude version could therefore be. (I know : Why csv ? Because a lot of person still use excel 😦 )

#Create a fake positions list
cat >Liste.csv <<EOF

#Create a ss directory
mkdir IMG
#Write the header of the script
cat >Batch.igv <<EOF
load igv_session.xml
snapshotDirectory IMG
#Now parse the csv file
gawk -F ";" -v R=10000 '{print "goto "$1":"$2 -R"-"$2 + R"\nsnapshot Screen"NR".png"}END{print "exit"}' Liste.csv >>Batch.igv

From, igv, go to Tools, Run a batch script. Load Batch.igv, when all the process will be done, IGV will terminate and you’ll find your screenshots in IMG.

For an even more automated version you can use the script “PrepIgvBatch.sh” available in the scriptotheque

More on polled

July 13, 2013 Leave a comment

iris.I’ve collected so far 23 answers to my little survey (by the way, a great “thank you” to all of you who answered). Just to remind to those of you who haven’t bet yet, why being polled and red carrier could be of interest for a breeder, here comes some R code. First you ‘ll have to download French’s holstein  gEBV here. Then start R in a terminal.

#Find file name (I suppose the file is unique and in the directory)


#Polled genotype===================================
#create a vector of status "pp=>Normal,Pp=>heterozygous polled,PP=>Homozygous polled"
#Identify heterozygous polled with grep applied to the name of the sire
PO[gEBV$NOTAUR[grep(" P | P$",gEBV$NOTAUR,perl=TRUE)]]="Pp"
#Identify homozygous polled
PO[gEBV$NOTAUR[grep(" PP | PP$",gEBV$NOTAUR,perl=TRUE)]]="PP"
#Draw a boxplot
boxplot(gEBV$X0 ~ PO ,col="blue")
#Add the Number of observation
for(i in 1:3){text(i,0.95*max(gEBV$X0),paste("N= ",table(PO)[i]))}

#=Now the Red factor================================
RF[gEBV$NOTAUR[grep(" RF | RF$",gEBV$NOTAUR,perl=TRUE)]]="Rr"
#Draw a boxplot
boxplot(gEBV$X0 ~ RF ,col="red")
#Add the Number of observation
for(i in 1:dim(table(RF))){text(i,0.95*max(gEBV$X0),paste("N= ",table(RF)[i]))}

So for a given genotype, chances to be in the “Elite” will vary a lot. Iris and her brother have pedigree (gEBV) of 155. Their parents reliability is .7.

mu=155 ; VarG=25 ; Rel=0.25*(0.7+0.7)
#Boxplot of gEBV depending on genotypes (in the actual sire population)
boxplot(gEBV$X0 ~ RF + PO ,col="grey",main="gEBV according to genotype")
#Add E(EBV) and trace a 90IC area

bpSo, we can clearly see that for Iris and her brother their career will mostly depend on the fact that they are red and polled carriers !

Categories: Agriculture, Funny science, R

Recent articles

July 8, 2013 Leave a comment

Just a small review of the articles I went through recently :

Sufficiently rare to be mentioned, Bayes theorem in Science  by  Efron , with a nice follow-up post on the og

Although the lab technicalities were far beyond my understanding, the questions raised by this article on Evolution of  essential gene, stroke me !

I wish I had time to have a look on these kind of procedure during my  Ph’D, a simple permutation algorithm to compute significance threshold. By the way, I also learned a new distribution : the Rademacher distribution

I was eager to see this article, the Rat Genome Sequencing and Mapping consortium  made a very interesting piece of work combining sequence and genetic mapping in outbred rats. A lot of questions came to my mind based on these results…yeah hunting the so called “causal mutation” may not be that easy.

And last the funny  article of the month ! This kind of question could have been seen on  Freakonometrics

Software update

June 18, 2013 Leave a comment

Another month, another software update

samtools 0.1.19

I missed it in the last posts (this version was released in March). Multi-thread is now available. I play a bit with the the different displays (Html, Curse and Text) of tview, (pretty handy). I may need more time to get acustomed to bamcheck/plot-bamcheck output.

Delly 0.0.11

Some small fixes, a new progress bar to control process, better temporary file handling, and BWA-mem support…more here

BWA 0.7.5a

Several small fixes

Picard tools 1.93

Small bug fixes. Released some hours ago (I had to compensate samtools glitch)

llvm 3.3

Still some efficiency enhancement, clang is now totally C++11 complient. New Arch support…and more. Release announcement is still pending, but binary are already available.


Apparently computing speed improvement are the newsworthy point of this release.

Coming soon…

I’ve had access to both next cuda 5.5 RC and Intel compiler….and so far the good new is that both now offer a simplified install procedure. Cuda 5.5 is available as a deb file, Intel compiler as a script with a GUI. I am eager to test these RC.