[jeannelouise]

Illumina HiSeq Inconsistent Reads

One more question relating to inconsistent read counts, have lab techs been a common source of error in processes such as library prep, quantification and machine loading? What about reagents for the Illumina? Has anyone experienced problems with those?

[Brian Bushnell]

Please create any future related posts in this thread rather than creating new ones.

[GenoMax]

@jennelouise: You had indicated in a previous post that you work at a core facility. You must not have been one of the people who were originally trained when the sequencer was installed. If you ask nicely your local FAS may be willing to come by and provide additional training. It sounds like you need someone to go over your specific datasets and provide some assistance.

While most of your question above can be generally answered in the affirmative they will have caveats associated with them. Many have already been covered in other threads here in different contexts.

Can you give concrete examples of what you consider as "inconsistent" reads? Is the inconsistency in terms of number or quality (alignments etc)?

[jwfoley]

Troubleshooting questions generally work best if they include information of the form "I did X, and I expected Y to happen, but Z happened instead."

In this case, that might be:

• I had some libraries made by [protocol, e.g. RNA-seq, RRBS, ChIP, exome]
• I quantified them by [method, e.g. spectrophotometry, fluorospectrometry, qPCR]
• I pooled them and loaded them into the flow cell at [molarity] according to this quantification
• In [some condition] I got [X] clusters but in [other condition] I got [Y] clusters [at least I assume you're talking about the number of clusters but even that isn't really clear]

People here are generally eager to help, but you're not making it very easy for us. To answer your question we'd basically have to write you a whole textbook, when maybe all you really need is someone to recognize a specific issue in one of your protocols, or just to tell you that the results you're seeing are actually normal.

Pending this information, I can certainly say that yes, the number of clusters you get depends on how accurately you quantified the libraries. Of course. qPCR is the gold standard and spectrophotometry is notoriously unreliable, while many people are happily compromising with fluorospectrometry and it works okay.

[pmiguel]

Actually, clustering on an Illumina flowcell is the gold standard. Yeah, it's crazy expensive so no one wants to use that as an assay, but we might as well face facts.

Our experience is that gel size selected, heavily amplified libraries give you the most accurate cluster titers. By which I mean you can use qPCR or even fluorimetry and nail cluster densities.

We generally are able to land in a reasonable cluster density range with any library using qPCR (KAPA kit on an ABI ONESTEP). But this does require periodic adjustment of the concentration of libraries we cluster. For example using qPCR on no-amp DNA libraries we were hitting the 900-1000 k-clusters/mm^2 cluster densities on our HiSeq 2500 at 20pM library concentrations. But over time we saw our densities fall when we clustered at this concentration and had to bump them up to 22.5pM. More recently we've had to back that number down a little.

No-amp libraries are the worst in this regard. Non-gel selected amplified libraries tend to be a little more stable.

As to why this is happening, we don't know. Could be fluctuations in library insert sizes, differences in clustering performance at different times of the year, differences in results from our qPCR machine, etc.

That is just part one, though. Hitting overall cluster densities on a lane is fairly trivial compared to getting even representation from libraries clustered together. Again, I think this works pretty well for gel-size-selected libraries that have been heavily amplified. But we basically don't make those any more. So we find that about 1/2 of the time we get at least one or two outliers in moderate to large size pools (say 6 samples or above).

At various points we've torn our hair out trying to figure out the source of this variation, but no joy. So we just do the best we can. Use qPCR for the individual libraries, pool according to our results there and then use qPCR to check the titer of the final pool. If the project involves more than one lane, we'll just put one lane on the first flowcell and adjust the pooling, if necessary, to compensate for very high or low cluster numbers for outlier libraries and correct for it in subsequent lanes.

Obviously this can delay returning results to customers so it isn't a great solution. But it is the only one we've found.

--
Phillip

[jeannelouise]

Currently, the Illumina machines are sometimes producing an inconsistent number of reads for pooled libraries from the same customer, at the same concentration, run in the same lane of a flow cell. The number of reads (approximately 125 million) should be evenly distributed between each library in that lane.

[jeannelouise]

No, we are not mixing samples from different submitters in one lane.

[HESmith]

Our solution is to require users to quantify the libraries and create the pools themselves. We provide the instrumentation (Qubit, Bioanalyzer, qPCR instruments), reagents, training, and assistance with troubleshooting, but ultimately it's his/her responsibility to pool at equimolar concentration. This strategy also moots the question of responsibility when pools contain the wrong libraries/indices.

[MU Core]

jeannelouise, as previously mentioned this could be a problem of expectations. Can you tell the group what would constitute an acceptable pooling. We find that pooling ratios with differences of 10% from the expected to be the norm. Sometimes you will find libraries as much as 20% from the expected due to previously mentioned variables.

[NextGenSeq]

Use QPCR to quantify your libraries prior to pooling.
The QPCR uses the same primers as those on the flowcell and it is the most sensitive and accurate method for quantification.

Do a search for KAPA Biosystems

[jeannelouise]

I am mixing samples from different submitters in one lane.

[GenoMax]

No, we are not mixing samples from different submitters in one lane.

I am mixing samples from different submitters in one lane.

[Jessica_L]

There's a lot of very useful information in the above posts already, but here are my thoughts:

I think the problem here is your expectation, jeannelouise. Even if you weren't mixing samples from different submitters, no two libraries will perform exactly the same. It's normal for a little fluctuation to occur- in the clustering process or some reads may get filtered out for quality reasons, etc.. The best way to deal with this scenario is, as others have said, to use qPCR for quatitation of your libraries and manage the expectations of your customers. If you're not sure if your results are reasonable, you can also share some of your quality metrics here-- I'm sure plenty of people would be happy to tell you whether your results are inline with what they observe on their own instruments.

On the other hand, it's a slightly different story if you're expecting an evenly balanced 15% across all of your libraries and some are 0 while others are 40%. Those kinds of issues usually represent systemic problems around normalizing library quantities* or a failure of the library prep itself. You would need to share a bit more information about the prep method-- I understand you may not have that information since you didn't prep some libraries yourself, but if you're a core facility, you may want to start requesting that info from the submitter. When I worked in a core facility, I frequently noticed submitters would send us libraries of sub-standard quality which would, of course, result in sub-standard data (and sometimes read counts). The way we dealt with that was by telling our customers that if they were going to prep the library themselves, they assumed all the risk if their samples failed provided that our in-run controls (i.e. spiked-in PhiX) performed normally.

I hope that helps.

---------------

* Seriously, get a qPCR machine and use that for quantifying libraries. Compared to Qbit, it's going to make your life a lot easier. I think Qbit is really only recommended for bisulfite treated DNA since qPCR data is unreliable for those libraries.