Insert size != Fragment size?

[Boel]

Hi All,

This is a very simple question, or should be, but there seems to be some confusion out there.

My understanding is that insert size is the number of bases between paired end reads (example: 300 bp fragments, 2*75 reads -> insert size 150). However, looking trough different threads here and information elsewhere the fragment is sometime referred to as insert as well, making insert size == fragment size.

I wonder whether different programs (BWA, picards CollectInsertSizeMetrics, bfast, samtools etc) have different definitions of insert size, which might make things messy. Right now I am especially interested in picards definition.

Any ideas?

Thanks,
Boel

[cw11]

I'm not sure about Picard specifically, but I found a thread that discusses insert size here, which seems to suggest that the insert size is the stretch of sequence between the adapters (so in your example, 300 would be correct). However, certain tuxedo suite programs (tophat/cufflinks/bowtie) take a --mean-inner-dist option, defined as fragment length - reads.

[nickloman]

Yes, I've seen both definitions used by different software. I know Bowtie uses insert size == fragment size for example.

Personally I like this definition because you could sequence the same library with different read lengths, or you could have variable length reads (Ion Torrent) or you could trim your 3' read tips. Each step would vary the insert size, but the fragment size would remain constant.

I think part of the difficulty comes from the difference between Illumina paired-end protocols (e.g. bidirectional sequencing) where insert size is always related to fragment size and the long mate-pair/jumping protocols, where the insert size relates instead to the sizing step (e.g. 8kb gel-cut) and is independent of fragment length.

[Boel]

According to samtools help (which includes picard):
"For Illumina paired-end data, the inferred insert size would be the difference between the 5' positions of the two reads." This translates to 300 bases in my previous example, since nucleotides are added in 5' to 3' direction.

Thanks for replying!

[cw11]

Yup - Glad you found an answer!

[jfostel]

Whether insert size = fragment size does vary from tool to tool, I would specifically look it up for whatever you're using.

Regardless, the total adaptor-insert-adaptor length is useful as the best predictor of a library's amplification behavior (both in qPCR QC and on the flowcell). For example, you wouldn't want to pool together two libraries with identical 200bp inserts but very different adaptor + index lengths (unless it was acceptable for the majority of the reads to come from the smaller construct).

[rskr]

Don't bother, use a different term. You will be misunderstood. When you mean fragment size, say "fragment size", when mean the distance between the pairs. Say, "distance between the pairs".