MiSeq lower quality on the bottom surface

M4TTN · 10-07-2015, 09:31 AM

Hello all,

Just wondering if anyone else that uses a MiSeq has observed this issue? We invariably see that the error rate and quality scores are worse on the bottom flow cell surface than the top.

In our current run, the difference is stark and will probably lead to the run failing to meet Illumina's quality standards.

Here is an example of some of the plots:
https://www.dropbox.com/s/90tcq29gyq...Run12.pdf?dl=0

Cluster density was 1430K/mm2 but with only 75% passing filter.
Kit is a 2x300 v3. Samples are NexteraXT from gDNA.

To me it looks like a problem with focussing on the bottom surface - leading to a failure to accurately pick out clusters and also subsequently call the sequences.

thermophile · 10-13-2015, 12:23 PM

look at your FWMH-that if it's a focusing issue it will show up there. The intensity is lower on the bottom-maybe your error rate is higher just because of the lower intensity.

M4TTN · 10-14-2015, 12:34 AM

Here are the FWHM plots:

https://www.dropbox.com/s/d0k73ecj0m...0FWHM.pdf?dl=0

The bottom surface is broader and spikier - but not sure if this is a significant (meaningful) difference.

By the way, final stats for the run were as follows:

Code:

 Run Summary
Level	Cycles	Yield	Proj	Aligned	Error	Intens	% >= Q30
Read 1	300	7.42	7.42	0.85	4.31	184	67.91
Read 2	8	0.17	0.17	0.00	0.00	469	90.70
Read 3	8	0.17	0.17	0.00	0.00	423	85.89
Read 4	300	7.42	7.42	0.78	6.07	171	54.65
NI Tot	600	14.85	14.85	0.81	5.19	178	61.28
Total	616	15.20	15.20	0.81	5.19	312	61.90

Lane	Tiles	Density (K/mm2)	Cluster PF (%)	Phas/Prephas (%)	Reads (M)	Reads PF (M)	% >= Q30	Yield (G)	Cycles Err Rated	Aligned (%)	Error Rate (%)	Error Rate 35 cycle (%)	Error Rate 75 cycle (%)	Error Rate 100 cycle (%)	Intensity Cycle 1
Read 1
1	38	1430 +/- 36	74.39 +/- 13.26	0.286 / 0.053	33.24	24.83	67.91	7.42	299	0.85 +/- 0.13	4.31 +/- 0.89	0.28 +/- 0.16	0.79 +/- 0.65	1.02 +/- 0.78	184 +/- 15
Read 2 (I)
1	38	1430 +/- 36	74.39 +/- 13.26	0.000 / 0.000	33.24	24.83	90.70	0.17	0	0.00 +/- 0.00	0.00 +/- 0.00	0.00 +/- 0.00	0.00 +/- 0.00	0.00 +/- 0.00	469 +/- 127
Read 3 (I)
1	38	1430 +/- 36	74.39 +/- 13.26	0.000 / 0.000	33.24	24.83	85.89	0.17	0	0.00 +/- 0.00	0.00 +/- 0.00	0.00 +/- 0.00	0.00 +/- 0.00	0.00 +/- 0.00	423 +/- 113
Read 4
1	38	1430 +/- 36	74.39 +/- 13.26	0.140 / 0.023	33.24	24.83	54.65	7.42	299	0.78 +/- 0.10	6.07 +/- 1.33	0.51 +/- 0.24	0.79 +/- 0.38	0.95 +/- 0.43	171 +/- 49

thermophile · 10-14-2015, 08:37 AM

I just looked through several of my runs (and my machine has focusing issues) and I don't see that much difference between the 2 surfaces. Have you sent this run to tech support yet? I think your q30 is below spec but run v2 500 cycles so don't know for sure the cutoffs for the 600 cycle kit.

M4TTN · 10-14-2015, 08:47 AM

Yes, we've contact tech support. They responded last week, but nothing since...

The run is definitely below spec. When using v3 chemistry, the published specs state 2x300 should generate %Q30 >70%

We've actually had a few runs replaced due to dropping below 70%, but I don't think any have had such a massive different between top and bottom of the flow cell.

Overall, our read4 is also always worse than read 1 (but I understand that is expected due to the turnaround step that broadens/merges clusters slightly).

What I don't really get, is that our first ever run gave extremely good data (Q30 > 80% I think), but we've not since been able to achieve that level even after 2-3 engineer visits.

GenoMax · 10-14-2015, 09:04 AM

Unless you are sequencing the same library (where you got 80+% Q30) there is no specific reason to get that PF everytime. That metric is going to be library quality dependent.

M4TTN · 10-14-2015, 09:14 AM

Indeed, these are different samples each time.

Any idea what might have caused the first library to be so much better?

We prepare genomic DNA by a similar method each time and prepare samples using NexteraXT.

Even when we get good clustering (25 M) and good % passing filter (>90%), the resulting Q-scores never hit what that first run gave us.

Perhaps it is simply variation in insert-length due to variation in the Nextera efficiency?

In any case, none of this explains (I don't think) the lower quality on the bottom surface. Something that happens from time to time - and when there is a difference it is always the bottom flow cell surface that is worse.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
WHY are the first few bases of Illumina HiSeq reads of lower quality?	NYGen	Illumina/Solexa	5	08-14-2015 09:04 PM
Lower quality in reverse reads only. Is it normal?	drramki	Bioinformatics	8	07-31-2014 04:58 PM
MiSeq quality of R2 reads	textux	Bioinformatics	2	05-14-2014 07:59 AM
Bottom Middle Swath and other focus issues	csquared	Illumina/Solexa	21	03-02-2012 06:19 AM

10-07-2015, 09:31 AM	#1
M4TTN Member Location: UK Join Date: Jan 2014 Posts: 75	MiSeq lower quality on the bottom surface Hello all, Just wondering if anyone else that uses a MiSeq has observed this issue? We invariably see that the error rate and quality scores are worse on the bottom flow cell surface than the top. In our current run, the difference is stark and will probably lead to the run failing to meet Illumina's quality standards. Here is an example of some of the plots: https://www.dropbox.com/s/90tcq29gyq...Run12.pdf?dl=0 Cluster density was 1430K/mm2 but with only 75% passing filter. Kit is a 2x300 v3. Samples are NexteraXT from gDNA. To me it looks like a problem with focussing on the bottom surface - leading to a failure to accurately pick out clusters and also subsequently call the sequences.

10-13-2015, 12:23 PM	#2
thermophile Senior Member Location: CT Join Date: Apr 2015 Posts: 243	look at your FWMH-that if it's a focusing issue it will show up there. The intensity is lower on the bottom-maybe your error rate is higher just because of the lower intensity.

10-14-2015, 08:37 AM	#4
thermophile Senior Member Location: CT Join Date: Apr 2015 Posts: 243	I just looked through several of my runs (and my machine has focusing issues) and I don't see that much difference between the 2 surfaces. Have you sent this run to tech support yet? I think your q30 is below spec but run v2 500 cycles so don't know for sure the cutoffs for the 600 cycle kit.

10-14-2015, 08:47 AM	#5
M4TTN Member Location: UK Join Date: Jan 2014 Posts: 75	Yes, we've contact tech support. They responded last week, but nothing since... The run is definitely below spec. When using v3 chemistry, the published specs state 2x300 should generate %Q30 >70% We've actually had a few runs replaced due to dropping below 70%, but I don't think any have had such a massive different between top and bottom of the flow cell. Overall, our read4 is also always worse than read 1 (but I understand that is expected due to the turnaround step that broadens/merges clusters slightly). What I don't really get, is that our first ever run gave extremely good data (Q30 > 80% I think), but we've not since been able to achieve that level even after 2-3 engineer visits.

10-14-2015, 09:04 AM	#6
GenoMax Senior Member Location: East Coast USA Join Date: Feb 2008 Posts: 7,083	Unless you are sequencing the same library (where you got 80+% Q30) there is no specific reason to get that PF everytime. That metric is going to be library quality dependent.

10-14-2015, 09:14 AM	#7
M4TTN Member Location: UK Join Date: Jan 2014 Posts: 75	Indeed, these are different samples each time. Any idea what might have caused the first library to be so much better? We prepare genomic DNA by a similar method each time and prepare samples using NexteraXT. Even when we get good clustering (25 M) and good % passing filter (>90%), the resulting Q-scores never hit what that first run gave us. Perhaps it is simply variation in insert-length due to variation in the Nextera efficiency? In any case, none of this explains (I don't think) the lower quality on the bottom surface. Something that happens from time to time - and when there is a difference it is always the bottom flow cell surface that is worse.