Re: [AMBER] Advice on Simulated Annealing

From: David A Case <>
Date: Wed, 29 Dec 2021 21:59:09 -0500

On Wed, Dec 29, 2021, Matthew Guberman-Pfeffer wrote:
>I’m wondering if advice can be offered on how to select the maximum
>temperature and heating/cooling schedules for simulated annealing.

If you don't have restraints on a protein, you probably need to keep the
maximum temperature pretty low (e.g. 300K): if your protein starts to unfold
(even locally) during the run, cooling is unlikely to bring it back to
a good structure.

If you can afford it, I'd suggest starting with something like 10 ns: say
run at 300K for 5 ns, and cool to zero over another 5 ns. Then see if you
get better/differnt results by doubling the time. (I use nmropt=1 to slowly
change TEMP0 from 300 to 0 during the second half of the run.

If you know where the strain is (e.g. between the porphyrin rings), you
could restrain parts of the system away from the problem area. Then you
could probably accommodate a higher maximum temperature. (You might want to
always restrain things that are far from the added porphyrins, on the
hypothesis that alphafold generally knows what it's doing.)

These are all guesses, based on limited experience, and others may
have their own anecdotes, or know of more systematic explorations. And
there are certainly many other accelerated sampling approaches that could be
used. But just letting the system rattle around for a while with MD will
often do wonders for localized strain.


Caveat: when I suggest protocols like the above to my students, it's pretty
common to find out that my advice wasn't very good.

AMBER mailing list
Received on Wed Dec 29 2021 - 19:00:02 PST
Custom Search