Hello,
I have Illumina paired-end/mate-pair reads and I would like to find out the min and max insert size is. For Aligner like SOAP2 is necessary to specify the insert size?
What is the best way to calculate the min and max insert size?
Thank you in advance.
From a SAM/BAM file, you can use Picard:
http://picard.sourceforge.net/command-line-overview.shtml#CollectInsertSizeMetrics
You can also get mean and standard deviation quickly from a SAM/BAM file using these two awk scripts, replacing YOUR_MEAN in the second line with the output of the first line:
head -10000 mappings.sam | awk '{if ($9 > 0) {S+=$9; T+=1}}END{print "Mean: " S/T}'
head -10000 mappings.sam | awk 'M=YOUR_MEAN{if ($9 > 0) {S+=($9-M)*($9-M); T+=1}}END{print "StdDev: " sqrt(S/T)}'
Or calculate both stats simultaneously:
awk '{ if ($9 > 0) { N+=1; S+=$9; S2+=$9*$9 }} END { M=S/N; print "n="N", mean="M", stdev="sqrt ((S2-M*M*N)/(N-1))}'
Note that this (and your examples) counts all pairs twice, you probably should check the flags to only count the "left" ends or something.
continuing the late-comment theme... For bowtie alignments, the TLEN field ($9 in awk) is negative for the reverse-strand end, so I don't think this will actually count twice with bowtie. This will, however, double count read pairs that align more than one place. You're right that you could do this better by bitmasking the FLAG field for first in pair and primary alignment, but not all versions of awk have this capability.
It gives strange results for mean, and I don't know why. For my sam file, for 10000 lines it gives 217 (picard gives 198). This script for all reads gives 1032,79. Overall it seems that the value increases along with the read number.
Edit.
Other, smaller, file. Picard mean = 156
Qualimap and mentioned awk script mean=1262.04
More strangely, the histogram of insert size has values ending at 475, median is 165.
If there are some outlier rejection in Picard? It looks to me that mean and sd are quite objective values.
•
written
by
boczniak767
I had the same question and helped myself by aligning my reads in several runs with some useful values for this parameter and calculating the insert size distribution afterwards with the script from this thread.
I had to estimate the parameter for a TopHat run and I tried several parameter values for the insert size and I found out that this parameter almost had no impact, at least on the dataset I used.
Hope that helped!
A bit late to the party, but I wrote a small tool to generate various statistics on insert sizes. It does require you to map to a reference, but the reference doesn't have to be very good as you only need enough matches to get an estimate.
More info here: http://blog.malde.org/posts/bamstats.html
Yes, I know. This was my primary motivation for doing quantiles instead, but I should probably do some outlier removal for the statistics. (Currently, my program streams over the data in linear time and constant space, and outlier removal will probably break that. :-( )
Log in to answer this question.
Usually it does not matter too much. Giving the max a large value (e.g. 1000bp for typical Illumina reads) is fine. Or use BWA, which estimates the insert size on the fly.