Speed up saturation scan

mtiberti commented 1 year ago

hi,

answering to your points:

re scaling with number of frames, yes, MutateX is fundamentally linear in this regard since each frame is treated independently
re scaling with system size, yes MutateX scales linearly respect to residue number, but I think FoldX itself scales worse than linearly for system size so your run could actually be quicker than just half the time our example run in the paper, since you have half residue number
the math about your system checks out, excluding other variables such as core speed or FoldX scaling as per previous points
yes, we do -n 5 by default. This comes from a note in some older FoldX documentation, in which it was reported that -n 1 would be ok for replacing a residue to most residue types, except those with long side chains with multiple Chi dihedrals (e.g. Lys, Arg) for which they recommended performing a indipendent runs to average out the effect of the modeling algorithm finding non-optimal solutions. We do report the standard deviation for the -n x runs in our results/pdb_name (second column of the files therein) in case you want to check that

aspitaleri commented 1 year ago

Thanks. mutatex performed well using -n 1, 299 aa in 24h using 32cores. Using -n 5 it is going slower, 69aa/day with 32 cores. I will compare the results between the two runs to have accuracy estimation between the two options.

aspitaleri commented 1 year ago

Just to complete the test. Here it is the plot n=1 vs n=5 using 50 frames. Small deviations is on larger DDG. The speed with n=1 is 4 times quicker than n=5 and considering the accuracy from this kind of calculation, it is fair to use n=1 with a bunch of frames.

mtiberti commented 1 year ago

interesting, thanks a lot! Definitely we should look into using n=1. Could you roughly quantify "small deviations" somewhat? From your plot is a bit difficult to tell especially at our range of interest (~ -3 to 5 kcal/mol), where also smallish deviations could change the classification of your mutation depending on what cut-offs you use

mtiberti commented 8 months ago

closing this issue for inactivity - thanks @aspitaleri for the useful comments :)

ELELAB / mutatex

Speed up saturation scan #154