[RASMB] averages from SV

[John Philo]

In my opinion Peter is placing too much emphasis on the importance of
absolute accuracy when calculating the averages over a distribution, and is
over-selling the advantages of his new c(s) approach. In truth ALL the
software approaches are only approximate, and what is important is simply
whether a given approach is adequate for a particular purpose. 

Further, while all of us our hopefully passionate about what we do, I think
we need to keep some perspective and should try to avoid trashing one
another's work in this forum, whether intentionally or not. I know of
roughly 70 publications using sedimentation coefficient averages to treat
reversible association, all derived using older methods, and I'm sure there
are many more out there. None of those publications actually happen to be
mine, but nonetheless I think it is getting out of line to imply that all of
those were based on poor methodology.

So now let me explain why I say c(s) is an approximation too and accuracy
may not be all that important. 

Remember, the c(s) approach is using a mixture model, which can never give a
perfect fit of the raw data for a dynamic, interacting system, and in
practice it certainly doesn't (there are always significant systematic
errors in the residuals). It is well known that c(s) often generates false
peaks, even when applied to samples that are truly mixtures, and false peaks
at the low or high sedimentation coefficient ends of the distribution (both
very common) will have particularly large consequences for computed Sn or Sz
values, respectively. False peaks are even more likely whenever the c(s)
model does not give a very good fit of the raw data, which as I said will
always tend to be true for the types of samples where these sedimentation
coefficient averages are most helpful. So while it probably is true that
c(s) will give more accurate values for the averages in many cases, the c(s)
is still approximate and not qualitatively different from the other methods.

Second, in reality for many purposes for which Sw, Sz etc. are used, the
absolute accuracy of the S values may not be terribly important. Remember
often one uses these exercises just to identify which models are worth
pursuing in more detail, by more rigorous approaches. Even when you are
fitting average values to a specific association model (for example to
obtain the energy of association), the S values for each species are simply
fitting parameters (or they are constrained to experimental values for pure
species that come from the same data analysis approach and thus have the
same systematic error). Thus some systematic error in those sedimentation
coefficients may not be significant, especially since in these complex
systems one is often quite happy to get an energy to +/- 1 kcal. 

John