Variants with extreme strand bias in HiSeq2500 data

sndrtj · 07-30-2014, 04:13 AM

We are doing whole exome sequencing on trios: an affected child and its parents, where we are looking for de novo variants that the child has but not the parents. The expected number of these de novo variants for each trio is just 1-3. Most our data comes from HiSeq2000 sequencing, but our sequencing center recently bought a HiSeq2500 machine. We tested a sample on the new HiSeq2500 (only the kids, we don't have HiSeq2500 yet for the parents), and overall everything looks much better. Higher coverages, increased base quality etc. However, where we previously found 1 de novo variant, we now find 27.

The 26 extra de novo variants are all transversions (T-->G & A-->C). When looking in the BAM file, we can furthermore see that all these 26 extra variants have 100% strand bias. That is: all the T-->G variants are found on the forward strand, whereas all the A-->C variants are found on the reverse strand. Since A-->C is the reverse complement of T-->G, I think I can make the assumption that all these 26 variants feature the same change (T-->G).

What is even more striking is that when I then fetch a region of 20bp around the variants from Ensembl, I can find a very clear, albeit small, motif: TCAT, where the first T is the affected base. This is unlikely to be something of our pipeline, since it is found in the very earliest BAM file (aka directly post-alignment).

Is there anything known about extreme stand bias with Illumina HiSeq2500, or can anything be done to prevent it? (Sure, I could filter it out, but that's not a very nice way of handling this imho).

EDIT: the chemistry type used was V4 for the HiSeq2500, V3 for HiSeq2000.

nucacidhunter · 07-30-2014, 09:45 AM

That is interesting. I wonder what was the chemistry type: V3, V4 or rapid and number of cycles.

sndrtj · 07-31-2014, 04:32 AM

V4 chemistry was used for the HiSeq2500.

GenoMax · 07-31-2014, 04:42 AM

You should report this via Illumina tech support to relevant internal development team at Illumina.

There may be a case specific explanation but if this is a general observation then the earlier it is confirmed the better.

nucacidhunter · 07-31-2014, 04:55 AM

If you have access to MyIllumina support bulletins, there is a bulletin about some anomalies about V4 chemistry in comparison to V3 released on 16/06 named "Important Technical Notes about HiSeq v4 reagents".

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
fundamental strand bias in sequence data	kingsalex	Bioinformatics	2	12-12-2013 07:52 AM
strand bias P-value	lyz1030	Bioinformatics	1	07-16-2012 01:12 AM
strand bias	zorph	Bioinformatics	0	06-13-2012 03:44 PM
Extreme nucleotide bias at fragment ends of Illumina mate pair library	kmcarr	Sample Prep / Library Generation	3	03-17-2011 02:03 PM
Regions of extreme bias in histone modification runs	greggrant	Epigenetics	8	05-27-2009 04:35 AM

07-30-2014, 04:13 AM	#1
sndrtj Junior Member Location: Netherlands Join Date: Jul 2014 Posts: 2	Variants with extreme strand bias in HiSeq2500 data We are doing whole exome sequencing on trios: an affected child and its parents, where we are looking for de novo variants that the child has but not the parents. The expected number of these de novo variants for each trio is just 1-3. Most our data comes from HiSeq2000 sequencing, but our sequencing center recently bought a HiSeq2500 machine. We tested a sample on the new HiSeq2500 (only the kids, we don't have HiSeq2500 yet for the parents), and overall everything looks much better. Higher coverages, increased base quality etc. However, where we previously found 1 de novo variant, we now find 27. The 26 extra de novo variants are all transversions (T-->G & A-->C). When looking in the BAM file, we can furthermore see that all these 26 extra variants have 100% strand bias. That is: all the T-->G variants are found on the forward strand, whereas all the A-->C variants are found on the reverse strand. Since A-->C is the reverse complement of T-->G, I think I can make the assumption that all these 26 variants feature the same change (T-->G). What is even more striking is that when I then fetch a region of 20bp around the variants from Ensembl, I can find a very clear, albeit small, motif: TCAT, where the first T is the affected base. This is unlikely to be something of our pipeline, since it is found in the very earliest BAM file (aka directly post-alignment). Is there anything known about extreme stand bias with Illumina HiSeq2500, or can anything be done to prevent it? (Sure, I could filter it out, but that's not a very nice way of handling this imho). EDIT: the chemistry type used was V4 for the HiSeq2500, V3 for HiSeq2000. Last edited by sndrtj; 07-31-2014 at 04:37 AM.

07-30-2014, 09:45 AM	#2
nucacidhunter Jafar Jabbari Location: Melbourne Join Date: Jan 2013 Posts: 1,238	That is interesting. I wonder what was the chemistry type: V3, V4 or rapid and number of cycles.

07-31-2014, 04:32 AM	#3
sndrtj Junior Member Location: Netherlands Join Date: Jul 2014 Posts: 2	V4 chemistry was used for the HiSeq2500.

07-31-2014, 04:42 AM	#4
GenoMax Senior Member Location: East Coast USA Join Date: Feb 2008 Posts: 7,080	You should report this via Illumina tech support to relevant internal development team at Illumina. There may be a case specific explanation but if this is a general observation then the earlier it is confirmed the better.

07-31-2014, 04:55 AM	#5
nucacidhunter Jafar Jabbari Location: Melbourne Join Date: Jan 2013 Posts: 1,238	If you have access to MyIllumina support bulletins, there is a bulletin about some anomalies about V4 chemistry in comparison to V3 released on 16/06 named "Important Technical Notes about HiSeq v4 reagents".