[RASMB] Photomultiplier tube disturbances?

[Borries Demeler]

> It would help tremendously to diagnose these issues if Beckman would just
> put the information about the PGA and PMT voltage for each scan into the
> .log files the GUI generates. That file is supposed to be for diagnostic
> purposes but it has very little useful information. 

I agree, that would be nice!

> But then it should also
> be trivial for Beckman to give us a mode specifically for pseudo-absorbance
> where the rescaling is suppressed. 

Yes, that would be a great improvement!

> I understand fully the advantages of pseudo-absorbance for lower stochastic
> noise and greater throughput. However to my knowledge no one has rigorously
> tested the advantages and drawbacks of the two modes, over a range of total
> OD, as you would do for qualifying and validating an analytical method in
> the pharmaceutical world. Just getting good residuals is not a test for
> either accuracy or precision.

Agreed, but what I showed were not just good residuals, but a systematic 
difference in magnitude of RMSD and that is most certainly desirable since
it gives you a better signal to noise ratio.

> In particular, the lower stochastic noise is
> clearly a much greater advantage when the absorbance is quite low, but one
> suspects the linearity (which is never very good in velocity acquisition
> modes) is even worse in pseudo-absorbance mode. If you run in the
> conventional way with water in the reference, record in intensity mode, and
> then analyze the data with 2DSA either as true absorbance, or as
> pseudo-absorbance, do you actually get the same answers (same within the
> estimated precision for each approach)? 

For the cases I have tried the answer is definitely yes. However,
absence of evidence is not evidence of absence. There could be the case
that proves me wrong. I must say that i haven't tried to go to very high
OD, but the problem there is the same, the pseudo-intensity residuals
are lower than the absorbance RMSDs. Furthermore, when you measure in
absorbance mode you convolute two different time invariant noise vectors,
the contributions of two different windows, which do not get subtracted
out. Generally they are lower than all the other contributors so the
magnitude of the TI noise generally is lower in absorbance experiments,
but it is clearly there and needs to be dealt with just the same.

> If the sample OD is near 1, do you
> actually get the same total concentration (same total OD of sedimenting
> species)? 

That depends on whether I run against water or buffer (which may absorb).
For that reason I always run a water channel for the reference. If there
is residual absorbance from the buffer, I want to know about it!

So when I fit the data from absorbance or intensity, the same relative concentrations
are generated (within noise levels).

> If the distribution details differ, which seems likely, how do you
> know which one is correct? If you do this for three replicates of the same
> sample what happens? If such studies have been published, I must have missed
> it.

I am not aware of such detailed studies having been published.

> The original paper on the pseudo-absorbance approach (Kar et al. 2000) did
> actually show differences in size distribution details [least-square g(s)]
> when such tests were made on a few samples, most of which I believe were at
> relatively high OD. 

...where linearity comes into question, so this could be the reason for
the differences.

> When you use pseudo-absorbance rather than true absorbance you give up
> something that may be important: zero absorbance has a unique physical
> meaning. Gaining a factor of the square root of 2 in stochastic noise is a
> good thing, but is that advantage ALWAYS dominant over the statistical
> drawbacks of introducing 50-100 additional fitting parameters (one per scan)
> to handle the RIN? Again, if this has been looked at systematically, I'd
> love to see that.

I haven't looked at this systematically, but can tell you that 1) in a well-functioning
experiment the RI noise component is actually quite small, close to random noise level.
2) the RI noise is easily linearly separable from the intrinsic signal and 3) the way
I correct for pseudoabsarbance is by always collecting the intensity profile of a 
water channel and using a DIFFERENT I_0 for each scan, based on the average intensity
of the water channel scan at this time. This takes care of most of the problems Troy
saw, but when the intensity suddenly jumps the difference in time may be too short
to provide the correct reference intensity for each scan all the time, so this is only
able to correct your data in a first approximation, the rest comes from RI fitting.

-borries