On Mon, 24 Jan 2022 at 22:49, Jessica M. Gonzalez-Delgado via
nmr-relax-users <[email protected]> wrote:
>
> Hi everyone,
>
> I was wondering if there is a way to know if the model-free-analysis is
> running properly or if it is stuck? I've run dauvergne_protocol.py on a
> cluster before for data of the same protein and it finished in less than
> 48h. I can see that the local_tm/ and sphere/ directories have been
> created, but the sphere directory only contains init/ and round_1/ and
> it's been 24h. The sphere/round_1/ directory contains directories aic/
> and m0/ through m9/, so I'm guessing it's done with round_1 but I don't
> know how to check. Thanks in advance!


Hi Jessica,

Welcome to the relax mailing lists!  Sorry for not replying to you
earlier.  I have been stuck at home for the last 2.5 weeks looking
after my 5 year old son due to local outbreaks of COVID-19 in his
kindergarten.  I'll respond to your other email here too:

> After a lot of troubleshooting it seems like the dauvergne_protocol.py
> only works when it is run with 5 processors. Is that right? I have
> access to a cluster and was trying to lower the calculation time by
> running it on more than 5 processors with no luck. I tried a variety of
> combinations (up to 59 processors) but 5 was the only one that allowed
> the calculation to not get stuck for days on round_1 of every model.

This is a clear sign that at least one input data point is bad.  The
model-free calculations on Gary Thompson's uni-processor and
multi-processor 'fabrics' will return the exact same result to machine
precision.  This is very thoroughly tested.  What you see with 5
processors tells me that you are getting stuck in the Monte Carlo
simulations for error analysis.  The relax logs should show this.
Monte Carlo simulations are random by design so that you do not get
the same result to machine precision ever.  Hence if you ran on 5
processors again, you might find yourself stuck.

The reason for being stuck on Monte Carlo simulations is the same in
all fields where it is used.  This is a common problem often seen by
those performing a model-free analysis with relax that normally
requires removal of bad input data.  It is not actually stuck, but
rather some or many of the simulation data points are very difficult
to optimise.  This has been seen a lot with relax due to it's very
high default accuracy and precision.  But it has always been due to
bad input data for one or more spin systems.  You should be able to
spot these in the relax log.  Then go back to your peak intensity
plots when you fit the R1 and R2 data and see if the curves for those
spin systems look reasonable.  The Grace graphs relax produces are
very useful for this data trimming process.  If the curve looks bad,
exclude that data.  I hope this helps.

Regards,

Edward


_______________________________________________
nmr-relax-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/nmr-relax-users

Reply via email to