[BioC] HTqPCR questions

[Kevin R. Coombes]

On 1/19/2011 5:40 AM, Heidi Dvinge wrote:
> Dear Alessandro,
>
> a long overdue answer...
>> Hi Heidi ! My name is Alessandro and i am a bioinformatician in a small
>> genome services company in Italy.
>>
>> We tested HTqpCR in a set of 20 ABI 7900 qPCR experiments with interesting
>> results (hence thank you so much for these libraries!!) but I do have a
>> couple of questions for you:
>>
>> - the library needs in input SDS output files, which contain the Ct
>> values called by the SDS software,
> Well, it just needs any sort of Ct values really. Many people get the
> output in SDS format, which is why I added an option for inputting these
> files directly. Technically, it can be any sort of test file, as long as
> it contains one ID-Ct value pair per row.
>> Now, it is possible to get these values automatically from SDS or
>> manually, trying to consider global
>> baseline and threshold settings for all the experiments. In your
>> experience, which is the most common
>> strategy for such kind of experiments which I would define medium scale ?
> I'm afraid I'm not 100% sure what you're asking here. It's been a long
> time since I tried playing with the SDS software. As far as I'm aware,
> most people tend to use the default settings for calculating the Ct
> values, but you can certainly adjust the baseline etc. within SDS if you
> ahve any reason for wanting to do so. As long as you keep the settings
> consistent across multiple plates you should be fine.
In my experience, it is usually better to use the SDS software to set 
*manual* baseline and threshold values.  Moreover, you will often need 
to set different baselines for different probes.  (For example, if you 
use ribosomal 18S as one of your control genes, the huge quantities of 
18S in the samples require a very early baseline, perhaps as small as 
1-3 cycles, and the interesting signal will often come off around 8 or 9 
cycles.  Many other genes will give better results with a baseline of 
around 3-12 cycles, with typical signals around 20 or more cycles. Even 
with the same baselines, you may need to use slightly different thresholds.)

The "quality" of the parameter settings, unfortunately, is typically 
assessed visually.  You'd like to see a pretty clear logistic curve, 
parallel for most samples, with the threshold cutting through the 
"linear" phase.  (I'd be a lot happier if I had some quantitative way to 
assess the results.)

One advantage of setting the parameter values manually instead of using 
the "automatic" setting in the SDS software is that the actual values 
used for the baseline and threshold will be available in the resulting 
data set, just in case you want to refer to them at some point in the 
analysis.  If you instead set it to "automatic", then the software does 
not report which values it chose for which probes.

I also find that you are better off quantifying as many of the plates 
(or cards if you are using ABI's "low density arrays") at once as the 
SDS software will allow, since that helps ensure that you use the same 
settings.

> Note though that this is not taken into account within HTqPCR. The package
> accepts whatever Ct values are provided as input as being "true". You can
> normalise for e.g. plate and batch effects, however there are no functions
> that will do any adjustment that depend on the initial parameters used for
> obtaining the Ct values. However, some of the normalisation methods are
> non-linear, i.e. they don't just subtract a constant from all Ct values,
> as the classical deltaCt method does. Hence, if you're not happy with the
> SDS software applies global settings to all samples within a plate, the
> normalisation can (at least partly) correct for that.
>
> I know that some people generally don't trust commercial software much,
> and prefer to get all the raw data, i.e. all the individual fluorescence
> measurements, and then fit some sort of sigmoid curve manually. However, I
> ahve yet to see any conclusive evidence that this is really necessary, and
> especially worth the extra effort. Perhaps other disagree with me here?
I spent some time playing with the raw data.  In part, I wanted to see 
what happened if you fit actual logistic models that account for the 
fact that some probe-primer pairs are probe limited and other are 
primer-limited. I decided that the basic model-fitting was good enough, 
in that accounting for these extra complications didn't seem to have any 
real payoff in the kinds of inferences one wanted to draw from the 
data.  None of this was published (since it's rather hard to publish 
something that says "this more detailed model with more parameters has 
no advantages over the current model for this kind of data"), and I'm 
not sure if I can even find the actual computations any more....

> HTH
> \Heidi
>
>> Kind regards,
>>
>> Alessandro G.