Re: [ccp4bb]: Refmac vs. cns

["Warren L. DeLano" <warren@sunesis.com>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***

On Mon, 21 Jan 2002, Anastassis "Tassos" Perrakis wrote:

> At resolutions better than ~2.0 there is not doubt in my personal (and 
> slightly biased, see also Warren/PyMol) opinion that ARP/wARP with 

Careful now, I'm also a CNS author : ), but I know that the real question
everyone is dying to have answered is, "How many levels of recursion can
you embed in a CNS or RefMac command file?"  Yes, CNS is a complete Turing
machine, and I don't think I can say the same for either RefMac or
ARP/wARP. 8-b

But seriously, I'll plug RefMac, ARP/wARP and the whole CCP4 suite over
CNS, not because they do or don't refine better at low resolution, but
because they are available to everyone (including industry) for a
reasonable price and *with* source code.  Relative performance hardly
matters if you can't justify the cost of the alternative package or can't
recompile it to run optimally on your system.  (HINT: If you want a job in
industry coming out of graduate school, then you are better off knowing
how to use cost-effective crystallographic software, such as CCP4).

By the way, simulated annealing as a general technique is only
guaranteed to converge if it is run infinitely slowly from very high
temperature.  Running it more quickly is bound to produce the local
optimization problems (errors) which have already been mentioned.  Those
are features of S.A., not bugs.  If you seek rapid, reliable convergence
for a specific class of problem, then there are theoretically more
efficient algorithms.  It shouldn't surprise anyone that other programs
can outperform S.A. -- in fact, it would be worrisome if they didn't,
since S.A. is essentially the simplist reasonable approach with a large
radius of convergence. 

Of course, *real* crystallographers don't mess around with those girly-man
refinement programs anyway.  They just glance at the intensities, do the
math in their heads, scribble down the new coordinates on a napkin, and
mail it in to the PDB.

There.  I didn't mention PyMOL once.             ......oops, darn!

- Warren