[nilshomer]

Would the original insert size for the library be listed in any resulting sequence alignment files (SAM/BAM for instance) for mate-paired reads (typicaly 2-5 kBp) or would it list the insert size of the paired-end reads (Typically ~500 bp). This would be important for the downstream Structural Variant analysis I can imagine...

I know that SAM/BAM is not (yet) produced by Illumina or Solid systems natively, but do any translators take care of this?

Thanks!

Take a look at the SAM specifcation. The insert size of an alignment (ISIZE) is defined as the distance between the 5' ends of the reads after mapping. The PI field in the RG tag in the header allows reads with the same expected insert size to be identified.

So to answer your question, if a read has its two ends 2-5Kbp or 500bp apart, ISIZE should be set accordingly. Optionally, the predicted size can be inferred from PI/RG in the header (when available).

[thondeboer]

Will original insert size of library be in SAM file header for Mate-paired sequences?

Would the original insert size for the library be listed in any resulting sequence alignment files (SAM/BAM for instance) for mate-paired reads (typicaly 2-5 kBp) or would it list the insert size of the paired-end reads (Typically ~500 bp). This would be important for the downstream Structural Variant analysis I can imagine...

I know that SAM/BAM is not (yet) produced by Illumina or Solid systems natively, but do any translators take care of this?

Thanks!

[ScottC]

Illumina refers to "paired end" as the original library preparation method they use, where you sequence each end of the same molecule. Because of the way the cluster generation technology works, it is limited to an inter-pair distance of ~300bp ( 200-600bp).

Illumina refers to "mate pairs" as sequences derived from their newer library prep method which is designed to provide paired sequences separated by a greater distance (between about 2 and 10kb). This method still actually only sequences the ends of ~400bp molecules, but this template is derived from both ends of a 2-10kb fragment that has had the middle section cut out and the 'internal' ends ligated in the middle. Basically, you take your 2-10kb random fragments, biotinylate the end, circularise them, shear the circles to ~400bp, capture biotinylated molecules, and then sequence those (they go into what is essentially a standard 'paired end' sample prep procedure).

[yvan.wenger]

paired-end and mate pairs

Paired-end and mate pairs. These two denominations refer to slightly different library preparations... correct?

Do we usually have the choice between the two? What are the practical differences between them? Anyone compared possible biases for example?

Best,

Yvan

[ECO]

That means?

Olivares is my surname as well. Haven't met many before.

[polivares]

Hey...Nice last name!

That means?

[ECO]

Hi all,
I am still confused with the difference between pair ends and mate pairs. Is it just that you call mates a pair of sequences (ends) from the same molecule?

Tnx

Hey...Nice last name!

[polivares]

Hi all,
I am still confused with the difference between pair ends and mate pairs. Is it just that you call mates a pair of sequences (ends) from the same molecule?

Tnx

[The_Roads]

no not yet. i'm going to have run through with them as soon as i can. thanks for the info

[nilshomer]

I am trying to quantify rare variants in deep coverage of small templates. I am not a statistician/bioinformatics pro but as far as i can see duplicate removal will introduce a bias that will enrich for rare variants both real and introduced.

Aside from library prep and pipeline issues which introduce their own biases, are there any assemblers that are designed for this type of assembly as opposed to large ref seq low coverage (<100x) assemblies?

Have you tried Velvet or Abyss? You can give either program the expected coverage and they will will work fine in my experiences.

[The_Roads]

I am trying to quantify rare variants in deep coverage of small templates. I am not a statistician/bioinformatics pro but as far as i can see duplicate removal will introduce a bias that will enrich for rare variants both real and introduced.

Aside from library prep and pipeline issues which introduce their own biases, are there any assemblers that are designed for this type of assembly as opposed to large ref seq low coverage (<100x) assemblies?

[nilshomer]

Thanks nilshomer, don't ask don't get, I should have come to SEQanswers earlier.

Anyone know of any other paired-end read generators?

Are there any with which you can model read errors, duplicate removal etc? or is this getting beyond their function.

Both come freely in the BFAST or MAQ distribution (I haven't checked other aligners).

I know the one I wrote (BFAST) models read errors both for Illumina or SOLiD, as well as SNPs and indels.

Why do you worry about duplicate removal? This can be frequent in practice in some cases.

[The_Roads]

Thanks nilshomer, don't ask don't get, I should have come to SEQanswers earlier.

Anyone know of any other paired-end read generators?

Are there any with which you can model read errors, duplicate removal etc? or is this getting beyond their function.

[nilshomer]

Hi mrwong05,

Synthetic read generator would be a very useful tool so i'll try and describe what we see with real world samples as best i can (and hopefully not air too much of our dirty laundry in public).

First some homemade terminology so you know what i mean, a paired-end run consists of two reads, 1 and its partner 2, and an unsequenced linker in the middle L. The read distance is 1+L+2.

When we do 2x 50 bp paired-end runs on a GAIIx using the current gel purification step we get read distances of between that vary by about 100 bp in a nice tight bell shaped curve starting between 160-200 bp. So the first thing to bear in mind is that L is not fixed within or between runs. Either way this group accounts for >99.99% of paired-end reads in an assembly. Because of the way fragments are generated for sequencing 1 and 2 can align either F-B of B-F.

If you want to be more realistic there are always a tiny proportion of reads <0.1% that align with much longer read distances, some of which is due to bioinformatics but some of which is real and simply reflects biology. Likewise a tiny proportion of reads at all read distances will be F-F or B-B. Also there appear to often be a tiny proportion of reads that come out overlapping where the read distance is the same as a read length ie 1+L+2 is, in this case, ~50-100. I have no idea of the prevelance of such reads but you can often find them if you look. Lastly if its not going to be part of the assembler, end trimming and quality trimming can often mean that 1 and 2 are different lengths and that a substantial number of reads from a paired end run end up with no partner at all.

I hope this is helpful. Please let me know how you get on with the read generator I would be very interested in using it to verify our sample analysis.

The_Roads

For example,b oth BFAST and MAQ have read generators, with BFAST having a paired end read generator for ABI and SOLiD data. Most aligner authors have their own read generators to validate and benchmark their aligners.

[The_Roads]

Hi mrwong05,

Synthetic read generator would be a very useful tool so i'll try and describe what we see with real world samples as best i can (and hopefully not air too much of our dirty laundry in public).

First some homemade terminology so you know what i mean, a paired-end run consists of two reads, 1 and its partner 2, and an unsequenced linker in the middle L. The read distance is 1+L+2.

When we do 2x 50 bp paired-end runs on a GAIIx using the current gel purification step we get read distances of between that vary by about 100 bp in a nice tight bell shaped curve starting between 160-200 bp. So the first thing to bear in mind is that L is not fixed within or between runs. Either way this group accounts for >99.99% of paired-end reads in an assembly. Because of the way fragments are generated for sequencing 1 and 2 can align either F-B of B-F.

If you want to be more realistic there are always a tiny proportion of reads <0.1% that align with much longer read distances, some of which is due to bioinformatics but some of which is real and simply reflects biology. Likewise a tiny proportion of reads at all read distances will be F-F or B-B. Also there appear to often be a tiny proportion of reads that come out overlapping where the read distance is the same as a read length ie 1+L+2 is, in this case, ~50-100. I have no idea of the prevelance of such reads but you can often find them if you look. Lastly if its not going to be part of the assembler, end trimming and quality trimming can often mean that 1 and 2 are different lengths and that a substantial number of reads from a paired end run end up with no partner at all.

I hope this is helpful. Please let me know how you get on with the read generator I would be very interested in using it to verify our sample analysis.

The_Roads