Closed GoogleCodeExporter closed 9 years ago
I have been testing g_rms a little more and noticed that it too sometimes
outputs similar non-zero values (10^-7nm = 10^-6 angstroms) when calculating
the RMSD of identical structures.
I was calculating the RMSD of a TM helix over a trajectory relative to its
starting position. Gromacs doesn't switch output from float to standard form
meaning I can't tell what is happening beyond 7 significant figures (not
normally an issue but one in this case).
Best,
Gareth
Original comment by MorganGa...@gmail.com
on 14 Nov 2012 at 3:34
Hi,
I made Josh owner of this ticket.
Just out of curiosity: I understand that the RMSD of a structure to itself
should be exactly 0 and this is what we would like to obtain ideally. However,
1e-5 Å is a fairly small number and in all applications I can think off I'd be
willing to count it as 0 (assuming that this is not a sign that the algorithm
itself is fundamentally flawed, which does not seem to be the case according to
the tests). I.e. if I calculated a RMSD of 1e-5 Å between two structures I'd
call them identical for all practical purposes. Why are you concerned with such
small differences?
Oliver
Original comment by orbeckst
on 24 Nov 2012 at 10:04
Hi Oliver,
My thinking at the time was 'if it can't produce the correct output in the one
known case (two identical structures), can I trust it with an unknown case?'
Subsequent investigation involving g_rms and reading the Theobald papers put my
mind a little more at ease - The QCP algorithm has a relative (to other
methods) precision of ~1e-6, which is in line with the discrepancies I've been
seeing.
Having investigated this I'm now happy enough to carry on using the MDAnalysis
RMS functions. Thank you both for your help on this.
Best,
Gareth
Original comment by MorganGa...@gmail.com
on 26 Nov 2012 at 2:59
Hi Gareth,
> 'if it can't produce the correct output in the one known case (two identical
structures), can I trust it with an unknown case?'
Makes sense.
> Subsequent investigation involving g_rms and reading the Theobald papers put
my mind a little more at ease - The QCP algorithm has a relative (to other
methods) precision of ~1e-6, which is in line with the discrepancies I've been
seeing.
Ok, then I'll close the ticket as 'WontFix'.
Thanks for reporting back.
Oliver
Original comment by orbeckst
on 26 Nov 2012 at 4:46
Original issue reported on code.google.com by
MorganGa...@gmail.com
on 14 Nov 2012 at 12:53Attachments: