Conventional semiempirical methods use matrix algebra methods, most of which scale as the third power of the number of atoms. A consequence of this is that modeling of systems of more than 1,000 atoms is impractical.

By using Localized Molecular Orbitals (LMO), the Self-Consistent Field equations can be solved in a time proportional to the size of the system. Increasing the speed of the SCF moved the slow step to other parts of the calculation, so changes were made to all the remaining time-consuming steps to make them more efficient. Changes were also made that reduced the memory demand. The result of all these changes is that routine modeling operations can now be carried out on over 90% of all the entries in the Protein Data Bank (PDB), previously only about 10% of all entries could be used.

Most of the common modeling functions have been enhanced by the MOZYME function, these include:

• Single-point calculations.

A general impression of the change in behavior can be obtained by comparing the resources required for various single-point, i.e., single SCF, calculations. Other operations, such as geometry optimization, consist of repeated SCF calculations. All calculations shown here were done using a Windows XP Pro computer with a 4000+ or 2.41 GHz Athlon processor.

The test systems used here were constructed from a single large protein, cut into pieces of the appropriate size.

1: Numbers in parentheses are estimated

After the first SCF calculation is done, all subsequent SCF calculations run much faster:

• Geometry optimization: This includes single point, restricted, and unrestricted, for molecular assemblies and for solids.

• Transition state location. This operation runs at about half the speed of geometry optimization.

• Transition state refinement. This operation runs at about 20% of the speed of geometry optimization.

• Transition state characterization. The FORCE calculation is a rapid operation now, running in a small fraction of the time required for geometry optimization.

• Intrinsic Reaction Coordinate. Unlike the other operations, this is very time consuming, taking 10 – 50 times longer than a geometry optimization.

• Reaction paths, grids, etc. These can be run using only the atoms of interest, in which case it is very fast, or using all atoms at every step, in which case it is slow. The time for each point is only slightly less than a geometry optimization.

For accuracy of modeling protein properties, see: HTTP<add URL here>

Utilities

• Identification of residues. The residue sequence can be calculated and printed. This allows unrecognized residues to be identified. If they represent user errors, then corrective action can be taken. If they represent genuine hetero groups, they can be re-defined as modified residues using keyword XENO. If they are the result of severe distortions generating spurious bonds, these bonds can be broken using keyword CVB.

• Identification of ionized residues. During a geometry optimization, hydrogen atoms might migrate to form a salt bridge. All resulting ionized residues are indicated by a trailing "+" or "­" sign. Partial charges on residues can also be printed.

• Ramachandran angles. All Ramachandran angles are printed when residues are identified.

• Atom and residue gradients. Total forces acting on atoms and on residues can be printed. This allows the rapid identification of geometric distortions, particularly in X-ray structures.

• Restart option. Because calculations involving proteins often take many days, a robust restart function is provided. During a long run, the restart file(s) can be used to run a zero-SCF, this allows a detailed examination of the state of the running system to be made, without stopping the run.

• Re-define chains and residue numbers

• Resequence. When hydrogen atoms are added, they are usually put at the end of the data set. This results in a non-standard PDB format. The standard PDB format can be generated using RESEQ. This puts all atoms in the same residue together. If two geometries, from entirely different PDB files are compared, the atoms in each file must first be put into the same sequence, so that there is an exact one-to-one correspondence. Unavoidable ambiguities in the PDB format definition require some extra re-arranging of atoms to ensure maximum coincidence. This is done using keyword GEO_REF.

Both because MOZYME is a new function in MOPAC, and because the issues involved in manipulating proteins are very different from those involving small molecules, all users are strongly recommended to spend some time familiarizing themselves with the new functions.

• Check hetero groups. Most utilities that add hydrogen atoms do a good job of handling the common residues, but a poorer job with hetero systems, such as the heme group. Because of this, almost all the errors in adding hydrogen atoms occur with hetero groups. Therefore, these groups should be checked carefully to verify that the degree of saturation is correct.

• Identify and characterize uncommon residues. If a residue is not recognized by MOPAC, check that it is correct. If it should be one of the 20 common residues, either add the missing atoms, or use XENO to allow the residue to be recognized. If it is a modified residue, such as the retinal group in bR, use XENO to re-define it as the parent residue  Do not ignore unrecognized residues – they usually indicate a fault in the data set.

To help with this, the following exercises are suggested:

1. Verify that MOZYME works: This is a simple example to show what a MOZYME output file looks like.  The system is Cys-Phe-Glu, but in the data set only the Cartesian coordinates are given.  By adding RESIDUES the various amino-acid residues in the tripeptide are identified. Information on the residues is printed shortly after the empirical formula is written in the output file. More new information on the residues in printed in the output.  This includes the total forces acting on the residues (a measure of how far the residue is from being optimized), and the net charge on each residues.  Most of these will be small, the exception being ionized residues, where the charge will be near to +1 or -1.
2. Generate a PDB file: MOPAC can convert a data set into PDB format.  The simplest way to do this is to add PDBOUT and 0SCF to the keyword list.  0SCF prevents any semiempirical calculations from being run, so this operation is very fast, even for large systems. If information on the secondary structure (helices and beta sheets, etc.) is available, add that information to the comments section of the data set (lines at the start of the data set that begin with an asterisk, "*").
3. Read in a PDB file: MOPAC should automatically recognize and read in PDB files.  All the text in the PDB file up to the first atom (a line starting with the word "ATOM") should be deleted, and the standard three lines of a MOPAC input data set put there instead.  If the PDB header text is needed, for example if PDBOUT is present, then convert the header lines into MOPAC comments by adding an asterisk ("*") to the start of each line.
4. Re-sequence a data set to conform to the PDB format:  If atoms are added or removed, so that the sequence of atoms in the data set does not follow that of the PDB format, the atoms can be re-sequenced, using keyword RESEQ.  This example is artificial, in that all atoms of each type are together.  When RESEQ is used, the job is stopped after resequencing.
5. Recognize an uncommon residue: In this example, a modified lysine group is correctly recognized by specifying the modification using  XENO.
6. Identify a fault in a data set: A data-set will not run because the charge specified (zero) is incorrect.  By using CHARGES, the ionized sites can be identified.  By analyzing these sites, any faults in the data set can be identified, and corrective action taken.  In this case, the only fault was an incorrect charge.
7. Optimize a polypeptide: In general, optimizing geometries is best done in two stages.  First, optimize the positions of all hydrogen atoms, then optimize the positions of all atoms.  In this example, all atoms are initially marked for optimization, but by switching off all optimization flags (NOOPT) then turning on the flags for hydrogen (OPT-H), only the hydrogen atoms are optimized. Once that's done, all atoms can then be optimized by switching on all the optimization flags (OPT).