Summary: TINKER vs CHARMm



 Hello,
 here is a summary of responses I received for my question I posted a
 few weeks ago.  Thank you very much for all the advice and comments
 Roy
 -----
 S. Roy Kimura
 Molecular Engineering Research Laboratory
 Department of Biomedical Engineering
 Boston University
 email: srk # - at - # bu.edu
 phone: (617)353-4697
 ----------------------< original question >------------------------
 Dear CCLers,
 Recently, I have been thinking about using the TINKER package (by Jay
 Ponder) rather than CHARMm since TINKER's source looks much easier to
 modify. As a check, I decided to calculate the internal energy of a 9
 residue peptide in vaccuum using both the CHARMm and TINKER
 programs. However, I have been getting rather different energies even
 though I used the same parameter sets and identical coordinates (for
 TINKER, I used the charmm.prm file supplied with the TINKER
 distribution which is supposed to be equivalent to CHARMm22
 parameters).  Does anyone have any idea what the problem might be?
 Roy
 -------------------------------------------------------------------
  Dear Dr. Kimura:
 I recently tested TINKER on a simple MD simulation of liquid argon.
 The calculated kinetic energy does not correspond to the temperature
 of the simulation. There are no obvious reasons for the error in
 the TINKER results).
 Hubert
 -------------------------------------------------------------------
 Roy,
 I've worked with a number of different force fields in different packages, so I
 can offer a few suggestions.  However, since I have not worked with either
 package, I cannot comment specifically.
 1) Check to see that the forcefields are defined exactly the same.  The program
 may say that it uses the same forcefield, but it could have been modified
 slightly.  This happens all the time with peptides, because everyone is
 interested in getting the best result.  (Ex. Improper torsional restraints may
 be used in one program to hold the peptide bond flat, but this may not be the
 case in the other program.)
 2) Check to see that the parameters are compatible as is.  In other words, check
 to make sure that they don't need to be divided by 2 or something simple like
 that.  Also, some programs make you enter the non-bonded parameters as sigma,
 epsilon, while others have you enter them a different way.  You said that you
 used a compatible parameter set, but is it possible that it came from a
 different version of the program?  Sometimes authors change these things while
 they continue to improve their software.
 3) Try a simple case like ethane or butane.  If you are still getting different
 values, then drop back to methane or similar without 1,4 interactions.
 4) Check to see that dielectric constants, H-bond functions, and such are turned
 on/off the same way in both programs.  Alos, the non-bonded cutoffs (if used)
 should be defined and set to the same value in both programs.
 5) Double precision vs. Single precision.  You said the coordinates were the
 same, but do the packages both use double precision throughout?  This would only
 account for very slight differences, but it is something to keep in mind.
 6) I really hate to say this, but if all else fails, trust the package that has
 been around the longest.
 Good luck.  Trust me, I know how frustrating it came be to find these errors.
 David Maxwell
 Senior Scientist
 Texas Biotechnology Corp.
 7000 Fannin, Suite 1920
 Houston, TX 77584
 E-Mail: dmaxwell # - at - # tbc.com
 Phone: (713) 796-8822 x102
 -------------------------------------------------------------------
  Dear Roy Kimura,
      I just returned from some time away from St. Louis and saw your
  post to CCL about "TINKER vs. CHARMm". I would be extremely
 interested
  in resolving (or at least rationalizing) the differences you are
  seeing between energies computed with our charmm.prm parameters and
  whatever CHARMm22 parameters you are using with "real" CHARMm.
      There are several possibilities. First, "charmm22" means
 different
  things to different people, MSI put out a version of charmm22 parameters,
  and Alex MacKerrell put out several versions over many years that were
  all loosely called "charmm22". The charmm.prm provided with TINKER is
 a
  translation to TINKER format of a version of charmm22 distributed with
  the related XPLOR package from Axel Brunger as of about mid-1994. At
  the time we compared energies from TINKER's charmm.prm with the
  corresponding XPLOR calculation, and got very close, but not
 "perfect"
  agreement for various polypeptide tests. The "official" paper on the
  protein parameters for charmm22 has just been published (J Phys Chem,
  102, 3586-3616, 1998). We certainly intend to update TINKER to this
  latest version of charmm22 and validate vs. the tests reported in this
  article, but we have not yet found time to do so. I would rather expect
  that the differences between the new "official" parameters and our
  older ones would be "small" (but not necessarily insignificant...).
      If you are instead getting "large" differences between TINKER and
  CHARMm, then the problem may lie elsewhere. Several people have caused
  themselves trouble by mixing parameter sets from one version of TINKER
  with source code and/or executables from another. This can lead to very
  odd results since we have frequently made changes to both parameter
  file formats and the code that are not "backward compatible".
 Regardless
  you should test with the most recent version of TINKER (currently 3.6).
  As a relatively new package, TINKER has undergone a large number of bug
  fixes as it has come into somewhat wider use. Also we have corrected
  mistranslations in our charmm.prm as they have cropped up (our charmm.prm
  was originally generated manually!). I know that some fairly early
  versions of TINKER were floating around BU Biomed. Engineering. Of course
  it is still possible that we have a "bug" in our implementation of
 the
  charmm function, but I tend to think we have things right at this point.
      Perhaps the best way to proceed, would be for you to send me the
  input and output files you are using for CHARMm (with as much detail
  as possible turned on for individual energy terms in the output). I'll
  take a look and see if I can't quickly find the origin of your trouble.
  Obviously, I'm very interested in seeing that TINKER reproduces charmm
  (and all the other force fields we support) as accurately as possible.
                                 Jay Ponder
 --------
 Jay W. Ponder				Phone:	(314) 362-4195
 Biochemistry, Box 8231         		Fax:	(314) 362-7183
 Washington University Medical School
 660 South Euclid Avenue			Email:	ponder # - at - # dasher.wustl.edu
 St. Louis, Missouri 63110  USA		WWW:	http://dasher.wustl.edu/
 -------------------------------------------------------------------
 THIS IS MY RESPONSE TO DR. PONDER'S POST
 Dear Professor Ponder,
 Since I posted that message, I experimented with different "charmm22"
 parameter sets, and found a set that gave charmm results very close to
 those from TINKER -- that is the charmm22 parameters that I downloaded
 from Axel Brunger's X-PLOR website (called top_all22_prot_b4.inp and
 par_all22_prot_b4.inp).
 In my initial test, I used the all atom "charmm22" parameters supplied
 with MSI, which, according to some people, is known as charmm19
 parameters, despite the fact that it says charmm22 in the beginning of
 the file.
 However, when I say very close, it means that it is much closer than
 what I had been seeing when I used the MSI parameters -- but there are
 some differences as you mentioned. Here are snippets from the outputs
 of both programs:
  ......................................................................
  Total Potential Energy :              -40.4963 Kcal/mole
  Energy Component Breakdown :         Kcal/mole      Interactions
  Bond Stretching                        18.2026           135
  Angle Bending                          23.3052           244
  Urey-Bradley                            1.8769           112
  Improper Dihedral                       0.1833            17
  Torsional Angle                        46.8472           362
  1-4 van der Waals                      89.1869           346
  Other van der Waals                   -10.2909          7921
  Charge-Charge                        -209.8075          7787
  ......................................................................
 ENER ENR:  Eval#     ENERgy      Delta-E         GRMS
 ENER INTERN:          BONDs       ANGLes       UREY-b    DIHEdrals    IMPRopers
 ENER EXTERN:        VDWaals         ELEC       HBONds          ASP         USER
  ----------       ---------    ---------    ---------    ---------    ---------
 ENER>        0    -36.49659      0.00000     10.32686
 ENER INTERN>       16.88385     26.02777      1.56513     38.97038
 0.18396
 ENER EXTERN>       91.72723   -211.85492      0.00000      0.00000
 0.00000
  ----------       ---------    ---------    ---------    ---------    ---------
  ......................................................................
 The above are the energy calculations from a single conformation of a
 9 residue peptide 2VAB (PHE ALA PRO GLY ASN TYR PRO ALA LEU). In both
 cases, I used the identical coordinates (same pdb file) which I
 created using charmm after minimizing the crystal structure by 50
 steps.  TINKER calculations were done with the latest version that I
 just recently downloaded (ver 3.6). In the TINKER keyfile, I have set
 the 1-4 scaling factor to 2.0 since that is what I understand charmm
 does with 1-4 nonbonded interactions. It appears the largest
 discrepancies lie in the torsional and Van der Waals energies (I
 presume one adds your '1-4' and 'other VdW' to get the equivalent of
 charmm's 'VDWaals'). Do you have any ideas as to why these
 discrepancies exist?