Conformational Searching Opcodes

The entire contents of the input file are used as the “seed” for the search for most search methods. At the start of a run, the full contents of this file is read, and the search proceeds as if all the structures in the input file had been found in the current search. This allows a new search to use the results of a previously run, partial search as its input. Of course, for a new search, the input file will contain a single structure.

The output file from the search contains in the header lines of the individual structures the number of times each structure was found. Therefore, the full benefits of usage-directedness are obtained in a subsequently run search. This degeneracy information is ignored, for technical reasons, during a network-distributed search. In a distributed search, or in any search seeded from an output file produced by an earlier version of MacroModel, the structures are treated as if they each occurred exactly once in a previous search. Debug (DEBG) flag 360, if set, instructs the program to ignore the degeneracy information even if it would otherwise utilize it.

Searches may be updated during program execution. When a search is updated, the information in the temporary file (filename.tmp) is written to the output file, after discarding structures that are not within the specified energetic window of the current global minimum. A summary of the job progress is also printed, which includes information on the number of times various structures were found and the convergence of these structures during minimization. Automatic, periodic updates can be controlled with the MCOP command.

For a discussion of conformational search protocols, see MacroModel Conformational Searches. As discussed there, it is advisable to perform the search without exhaustive minimization of all found structures and to subsequently reminimize the surviving structures.

This can be done without initiating a new MacroModel job, by inserting into the .com file a RWND command following the MINI command that terminates the search setup, then inserting a BGIN/END loop within which a READ/MINI sequence is specified. The COMP atoms used in the search will continue to be active during the reminimization.

The maximum number of conformations that can be stored when using MCMM, LMCS, SPMC, or MULT is no longer limited to 10000. See the respective command descriptions and MacroModel Program Capacity for further details.

The opcodes in this section are linked below:

AUTO — AUTOmatic setup

AUOP — AUto OPtions

LMCS — Low-Mode Conformational Search

LMC2 — Low-Mode Conformational search for large molecules

LOOP — protein LOOP conformation generation

LPOP — LooP OPtions

ARPK — ARPacK parameters

MCMM — Monte Carlo Multiple Minimum

SPMC — Systematic Pseudo-Monte Carlo search

MCOP — Monte Carlo OPtions

MCSS — Monte Carlo Structure Selection

FLAP — use FLAP algorithm for ring conformations

MCRC — Monte Carlo Ring Conformations

MCTS — Monte Carlo Torsion Selection

MCSM — Monte Carlo Single Minimum

MCNV — Monte Carlo Number of Variables

SEED — random number generator SEED

TORS — variable TORSion selection

TRES — Torsional RESolution

MOLS — variable MOLeculeS selection

LIGB — LIGand Bonds

RCA4 — Ring Closure Atoms

DISC — DIStance Check

TORC — TORsional Check

MCMF — Monte Carlo Maximum constraint Failures

SMPL — monte carlo SaMPLing

SRNO— SeRial Number Ordering

AUTO — AUTOmatic setup

Perform automatic setup for MCMM, MINI, SPMC, LMCS, LMC2, and combinations of MCMM with LMCS or LMC2, on the current structure. This procedure takes into account the current substructure information. The setup for MCMM is compatible with MCMM/LMCS and MCMM/LMC2. In addition, it can be applied in a serial manner (across different molecules in the input structure file) in MCMM, MCMM/LMCS, and MCMM/LMC2 serial searches, but not in pure LMC2 serial searches.

AUTO is incompatible with LOOP and must be used carefully with MBAE. In addition, AUTO is incompatible with frozen atoms unless a substructure is being used.

Note: Serial searches can generate large numbers of output structures, and the resulting output structure file can be large enough to exceed file size limits.

AUTO should be preceded by the MSYM, MCMM, MCNV, MCOP, LMCS, LMC2, SPMC, SUBS, and READ opcodes, if present, and is best placed just before the MINI line in the .com file.

The type of calculation in use is detected automatically. For MINI calculations AUTO sets up comparison atoms, chiral atoms, and torsional constraints unless instructed not to. For MCMM calculations AUTO sets up comparison atoms, chiral atoms, torsional constraints, molecule moves, variable torsions, and ring closures. For LMCS and LMC2 calculations, molecule moves, variable torsions, and ring closures are set up but not used, unless the setup is turned off. Series of bonded atoms that connect fixed or frozen atoms in substructures have ring closure bonds created within them to prevent rotation of fixed or frozen regions. Failure to find such a ring closure turns off all variable torsions in the SUBS atoms connecting such regions.

AUTO arg8 controls the extent to which variable torsions are applied to the input structures. This option is relevant mostly to amide and ester linkages, and supplies a greater degree of control over sampling of standard and nonstandard amides and esters. arg8=1 applies torsion moves to the amide/ester linkage of nonstandard groups like anhydrides, carbamates, hydrazones, and so on. Normal esters and amides are not sampled with this setting, but a torsional constraint is applied to them to ensure that the relative conformation of these groups is maintained. arg8=2 adds torsion moves to sample normal esters and amides in addition to amide and ester linkages of nonstandard groups. arg8=3 adds torsion moves to C=N and N=N bonds, in addition to all amide and ester derivatives. arg8=4 in addition removes torsional constraints involving hydrogen atoms.

In automatic setup, the maximum number of torsions varied at the same time is set to the minimum of arg2 of MCNV and the number of torsions present in the current system. If MCNV arg2 has not been set then the number of torsions present in the current system is used. In serial calculations, the number of torsions varied at the same time is determined for each system.

Related DEBG flags: 520, 521.

AUOP — AUto OPtions

When AUTO is used in conjunction with an MCMM, mixed MCMM/LMCS, mixed MCMM/LMC2, SPMC, or CGEN conformational search, AUOP may be used to modify the number of steps depending on some measure of the number of degrees of freedom present. AUOP should be used immediately before the AUTO command.

For CGEN by default the number of moves can also depend on ConfGen’s internal estimate for the number of degrees of freedom. See CGO2 arg3 for more information.

LMCS — Low-Mode Conformational Search

This method is described by Kolossváry and Guida [30, 31]. In tests so far, it has proved highly efficient, and it has the advantage that ring structures and variable torsion angles do not have to be specified. LMCS works by exploring the low-frequency eigenvectors of the system, which are expected to follow “soft” degrees of freedom, such as torsions. It is, however, helpful to specify independently movable molecules, by means of MOLS commands. Like the MCMM procedure, LMCS may be seeded or restarted with multiple input structures, and LMCS may also be used with distributed MacroModel. Most features that work with MCMM also work with LMCS. COMP, CHIG, DEMX, and MCSS commands should ordinarily be specified, and several MCOP arguments take on special meaning when used with LMCS.

The AUTO opcode can be used for automatic setup of LMCS and MCMM/LMCS calculations.

LMCS comes in two varieties: LMCS-SHAKE and mode-following. LMCS-SHAKE follows the low-mode vector, but every 2 Å it applies a few steps of steepest descent minimization to relieve any strain due to distorted bond lengths and bond angles introduced by the move.

The mode-following (eigenvector-following) option allows the user to apply the original low-mode search concept: “Since the potential energy hypersurface is a network of interconnected minima and saddle points, we reasoned that one could utilize a procedure that relies on eigenvector following for conformational searching. Thus, one could initiate the search by starting with any local minimum. By using one of the eigenvector-following techniques, one could locate a saddle point associated with this minimum and then the other minimum associated with this saddle point. By application of the eigenvector-following technique to the second minimum or to a different eigenvector of the first minimum, additional minima could be located which could then be used to find additional saddles, etc.” (quoted from the J. Comp. Chem. article cited above).

Mode-following uses the same eigenvector-following saddle point search method as in the SDLP command (see details there). If LMCS mode-following locates a saddle point, it proceeds as follows. The eigenvector for the negative eigenvalue is computed, followed by a short move along that eigenvector away from the starting structure. The resulting structure, which is slightly lower in energy than the saddle point structure, is then energy-minimized to locate the minimum energy structure on the other side of the saddle point.

If LMCS mode-following cannot locate a saddle point, the starting (minimum) structure is restored and is automatically “rejected by starting geometry.” Thus, only minima found via saddle points are stored. This makes LMCS mode-following useful for the mapping of conformational interconversions in a local region of the potential energy surface. The recommended procedure for this is to set LMCS arg5=−1 (every step mode-following), and MCSS arg1=0 (random walk structure selection) or LMCS arg4=1 (local search mode).

Note: LMCS mode-following is not intended for general conformational searches. Simple LMCS or LMCS-SHAKE (LMCS arg5>=0) is much more efficient for this purpose. LMCS mode-following is only recommended for local searches exploring the conformational interconversions of a molecule. DEBG 920 saves all the intermediate structures during LMCS-SHAKE or mode-following in the output structure file and colors them so that you can conveniently visualize the different multiple LMCS moves.

LMCS can be combined with MCMM search. Tests have confirmed that the most efficient use of LMCS is to allow explicit torsional rotation of key torsion bonds, especially in acyclic structures. Therefore, you can combine LMCS with TORS commands (with RCA4 if necessary) without restriction. MOLS commands can also be applied to translate/rotate independently movable molecules. See also the MCOP command.

LMC2 — Low-Mode Conformational search for large molecules

This method, termed LLMOD, is described by Kolossváry and Keseru [32]. LLMOD has been developed for large-scale conformational searching, such as protein loop optimization, homology model refinement, and fully flexible docking for induced fit modeling, only to mention a few applications [33]. LLMOD is very similar to LMOD, which is implemented in MacroModel as the LMCS command, but LLMOD utilizes the ARPACK package to compute low-mode eigenvectors of a Hessian matrix that is only referenced implicitly, through its product with a series of vectors. The Hessian × vector product can be calculated by a number of methods. LLMOD is the first conformational search method that can be applied to fully flexible, unconstrained protein structures.

LMC2 is implemented to extend the basic use of LMCS to molecules of virtually any size. Note that two main features of LMCS, LMCS-SHAKE and LMCS mode-following, are not implemented in LMC2. The command arguments of LMC2 are fairly similar to those of LMCS. Like LMCS, LMC2 can also be used in conjunction with the MCMM command to explicitly alter key torsion bonds via TORS commands (with RCA4 to open macrocyclic structures if necessary) and apply explicit translation/rotation of a ligand with the MOLS command for docking. LMC2 is primarily designed for fully flexible molecules, but the use of frozen atoms is allowed, and recommended in cases where parts of a macromolecule can be treated rigidly. Also note that you can set certain parameters controlling the ARPACK package via the ARPK command.

The AUTO opcode can be used for automatic setup of LMC2 and MCMM/LMC2 calculations. It can be used for MCMM/LMC2 serial searches but not for pure LMC2 serial searches.

LMCS is more efficient than LMC2 when the number of atoms is small enough for the entire Hessian to be stored in computer memory without swapping.

Note: It has proven to be good practice and time-saving to minimize molecular structures during a conformational search to a gradient that is not too low and only re-minimize entirely the final set of low-energy conformations. In that respect, LMC2 needs special attention. It is still recommended to follow this scheme, e.g., for unconstrained proteins, a gradient RMS of 1 kJ/mol/Å is sufficient during the search, but the very first structure, for which the low-modes are computed, must be pre-minimized to a low, < 0.1 kJ/mol/Å gradient RMS in order to derive reasonable modes. Pre-minimization can either be done in a separate MacroModel job, or the RWND command can be used to fully minimize the first structure, rewind the output file and start the LMC2 conformational search from there.

Note: Large-scale low-mode calculations can use a large amount of memory. If memory issues are encountered, consider eliminating the use of solvation. In addition, reducing non-bonded cutoffs also reduces the memory required for the computation.

LOOP — protein LOOP conformation generation

LOOP is a conformational search method for protein loops that works in conjunction with the MINI opcode in a manner similar to MCMM.

Given a single protein structure LOOP attempts to generate candidate loop conformations using either the specified loop from the protein structure provided or an alternate user specified sequence for the loop. LOOP rapidly generates a candidate structure with the appropriate bond lengths and angles. In addition, heavy atoms in this structure (i.e., atoms heavier than hydrogen) will not approach each other more closely than a specified distance. These candidate structures are then energy minimized and can be checked for uniqueness in a manner analogous to that for MCMM. A more general discussion of the use of LOOP is available in Protein Loop Construction with MacroModel.

LOOP generates candidate structures in two stages:

• Initial loop generation via random selection of rotatable dihedral angles without concern for loop closure

• “Tweak,” an iterative process involving the systematic adjustment of the rotatable bonds to attain loop closure [34, 35] while eliminating very close encounters amongst heavy atoms.

In the loop generation stage, the N-terminus and C-terminus halves of the loop are constructed separately, starting from their attachment points, by growing the chain out to the next rotatable bond. Then a random dihedral angle is chosen for this bond and the heavy atoms positioned unambiguously as a result are checked for close approaches with heavy atoms in the rest of the protein (not the loop itself). If there are such close approaches, the process of selecting a new random angle and checking for close approaches is repeated until either N_try such attempts are made or no close approaches result. If N_try such attempts fail then the loop construction process retreats back to the previous rotatable bond for which a new random angle is chosen. When a dihedral angle has been chosen successfully, the process focuses on the next rotatable bond out from the attachment point. Rotatable bonds in side chains are also sampled in this manner. The loop generation stage produces a random structure for the protein loop that is not connected in the middle and does not have close approaches between the loop atoms and the rest of the protein, but there may such close approaches within the loop itself.

In the second stage the Tweak algorithm is used to bring the middle of the loop together in the right geometry. Briefly, this algorithm proceeds by iteratively selecting a set of minimal changes to the rotatable bonds that lead toward loop closure based upon a linear approximation. Loop closure is specified by a set of target distances that must be met to within a tolerance of 0.001 Angstroms. Along the way close approaches involving heavy atoms of the loop atoms and the rest of the protein are removed by temporarily introducing additional target distances between the overlapping atoms. When the set of distances for closing the loop are met to within 1 Angstrom the Tweak algorithm also attempts to remove close approaches amongst the heavy atoms within the loop. By using a minimal set of changes to the rotatable bonds the Tweak algorithm retains as much of the original, random loop conformation in torsional space as possible and thus provides a mechanism for generating a diverse set of candidate loop structures.

Limitations:

• Only one loop can be treated, although successive calculations may target different loops.

• An all-atom representation of the protein must be used.

• The protein loop and the residues that it is immediately attached to must consist of alpha amino acids.

• Rings (e.g., the one in proline) are treated as rigid.

• No disulfide bonds are permitted within the loop or between the loop and the rest of the protein.

• No atoms in the loop can be frozen.

Using LOOP usually results in a renumbering of the atoms within the structure such that the loop atoms become the highest numbered atoms in the structure. LOOP tries to make this work by automatically shifting the atom numbers provided in SUBS, FXAT, FXDI, FXBA, FXTA, COMP, and CHIG commands. There are also routines for automatically generating COMP and CHIG atoms for the loop atoms as described for arg3, below. To facilitate subsequent studies using the structures generated, a new substructure file, out_filename.sbc, containing the shifted atom numbers is generated automatically. In addition, shifted COMP and CHIG commands suitable for use in .com files are written in the .log file.

If comparison atoms (COMP) are specified explicitly or implicitly (see arg 3) then MSYM must be used.

Related DEBG flags: 555

LPOP — LooP OPtions

LPOP provides a means to modify some of the parameters related to protein loop generation with the LOOP opcode. If no LPOP commands are provided the default values for the parameters are used. Multiple LPOP commands (e.g., with different arg1 values) may be used. It is not expected that these parameters would need to be modified from their default values under normal conditions so typically one would not need to include a LPOP command in the .com file.

Related DEBG flags: 555

ARPK — ARPacK parameters

This opcode applies to LMC2, VBR2, and LTNCG (MINI arg1=11)

The ARPACK package was developed by Danny Sorensen, Richard Lehoucq, Chao Yang, and Kristi Maschhoff at Rice University [36, 37]. Instead of attempting to solve a huge eigenproblem directly, ARPACK solves it indirectly by solving a series of small problems of much lower dimensionality and references the huge matrix only implicitly through its product with some vectors. ARPACK is ideally suited for LLMOD calculations, because direct calculation of the low-mode eigenvectors of a huge Hessian matrix of a protein molecule is prohibitive. Instead, LLMOD utilizes ARPACK to compute the low-mode eigenvectors indirectly. The ARPK command allows a user to fine tune ARPACK parameters. Note that the VBR2 command also uses ARPACK to calculate vibrational modes.

DEBG 999 can be used to provide greater verbosity for ARPK.

MCMM — Monte Carlo Multiple Minimum

MCMM [38, 39] is our recommended conformational search method. The input structure will be modified by random changes in torsion angles and/or molecular position as specified by the TORS or MOLS commands. Ordinarily, whether a single structure or multiple structures appear in the input file, they will first all be read in, minimized and treated as if already found by the MCMM procedure. This allows a new search to be initialized from the output of a previous search, by using the output file of the old search as input for the new one. However, if the necessary READ and MINI commands are placed within a BGIN/END loop, then a separate search is carried out for each input structure. This is called a serial search. For such serial calculations, the AUTO opcode is also needed.

The TORS command is used to specify dihedral angles to be varied; the MOLS command specifies relative positions of multiple molecules, as in an enzyme-substrate docking procedure. In addition, RCA4 commands and LIGB commands can be used to open rings and break ligand bonds, respectively, before performing torsional or relative molecular motion. CHIG commands should be specified for ring-closure atoms as well as to retain chirality about other centers. TORC commands may be used to hold double-bond configurations constant.

The AUTO opcode can be used for automatic setup of any kind of MCMM calculation.

Only unique structures will be retained, as in the MULT conformational searches. Use DEMX to set an energetic window to select low energy conformations. It is usually found that not all structures converge in minimization during a conformational search. A MULT minimization of the output file is recommended to achieve convergence for the final result.

While the default search method is random walk, we find that the usage-directed search (MCSS arg1=2) gives improved search performance.

SPMC — Systematic Pseudo-Monte Carlo search

Similar to MCMM, but invokes systematic search in place of random search. The search begins at low torsional resolution (120°), searches all angles without duplicating coverage, then doubles the resolution, etc. This method has the advantage of not retracing its path and consequently converges the final stages of the conformational search more efficiently than MCMM. Like MCMM, the method is effectively open-ended: it will search conformational space until stopped by the user or with arg1.

SPMC searches are conducted starting from a single input structure. However, if the necessary READ and MINI opcodes are placed within a BGIN/END loop, then a separate search is carried out ion each input structure. This is called a serial search. AUTO may be used to set up single SPMC searches and is required for serial SPMC searches.

It is suggested that torsional memory (MCSS arg3) be activated when using SPMC to prevent retracing of points in conformational space when starting from different starting geometries. If rings are being varied (i.e., RCA4 commands are being used), geometrical preoptimization (MCOP arg2) should also be activated.

Use with MCNV arg1=1 and arg2=N−1, where N is the number of variable torsions plus the number of molecules being independently translated/rotated with MOLS commands.

Details of the method are provided by Goodman and Still [40].

The arguments are the same as for MCMM, except arg3:

MCOP — Monte Carlo OPtions

This command alters the data written to the .log file, and also the geometrical optimization routine. If MCOP is omitted, this is equivalent to setting arg1=250 and arg2=0.

Note: Despite its name, MCOP specifies parameters for use in low-mode (LMCS and LMC2) and LOOP calculations, as well as Monte Carlo searches. Arguments 4 and 5 only apply to LMCS and LMC2 jobs. Starting in MacroModel 6.5, these arguments have new meanings and arguments 6 and 7, which pertained to the earlier LMCS methodology, have been eliminated.

Note: arg4=0 is unsuitable for non-conformers.

MCSS — Monte Carlo Structure Selection

This command allows the program to select starting geometries for Monte Carlo search steps in several different ways. Arg1 selects between random walk and two usage-related criteria. Arg2 affects weighting among selected structures for low energy geometries. Arg5 gives an optional energy window which prevents structures which are high in energy from being chosen.

This command is active only when doing global searching (i.e., when MCMM or SPMC arg4 = 0). In any case, structures must be within 100 kJ of the current global minimum to be candidates for starting geometry selection.

FLAP — use FLAP algorithm for ring conformations

Use the FLAP ring vertex reflection algorithm [54] to generate ring conformations in MCMM, low-mode, or mixed MCMM/low-mode conformational searches. This opcode must be added before the relevant MCMM or LMCS opcode.

MCRC — Monte Carlo Ring Conformations

Use templates of ring conformations in MCMM, low-mode, or mixed MCMM/low-mode conformational searches. This opcode must be added before the relevant MCMM or LMCS opcode.

MCTS — Monte Carlo Torsion Selection

Experience has shown that this command is not terribly useful.

Use in conjunction with MCMM to favor torsion angle selection near local torsional minima for the angles being varied. One MCTS command is required for each torsion which is to be effected. Arg1-4 are the atoms defining a particular torsion angle (i.e., arg2-3 should appear as a rotatable bond in a TORS command). Arg5-7 are 1-fold, 2-fold, and 3-fold torsional barriers which are used to compute local torsional energies as part of the test for an allowable value of the randomly selected Monte Carlo angular change. After a Monte Carlo variation of a torsion angle described by arg1-4 of this command is performed, a local torsional energy (E_T) is computed based on the value of the dihedral angle, τ , using the usual MM2 formula:

E_T = (V₁/2)(1+cos τ) + (V₂/2)(1−cos 2τ ) + (V₃/2)(1+cos 3τ ) − E_T,min

where E_T,min is the minimum possible value of E_T. If E_T is greater than 1, the torsion is rejected and a new random torsion angle is chosen. Otherwise, E_T is compared with a random number between 0 and 1 and, if E_T is larger than that number, a new random torsion angle is chosen. This scheme selects for local torsions which are low in energy. For sp3-sp3 linkages, one can favor the gauche and anti conformers (the minima) by using a V₁ (arg5) and V₃ (arg7)of 0.25. This will cause totally eclipsed (0-degree torsion) torsions to be strongly disfavored, 120-degree torsions to be moderately disfavored and the gauche and anti conformations to be favored. By choosing V₁-V₃ with care and using a random number for comparison, even high energy geometries are occasionally explored.

MCSM — Monte Carlo Single Minimum

This procedure is similar to that described by Li and Scheraga [41] and is a global search for the single lowest energy structure. The search can be conducted with or without minimization (depending on arg3). While this method is good at finding a single low energy conformer, there is no guarantee that it will locate the true global minimum energy conformer.

Monte Carlo internal coordinate conformational search is performed with minimizations on a single structure. Variable internal coordinates are specified by TORS and/or MOLS.

MCSM can be used with cyclic structures providing that either ring bonds are not varied or that ring closure commands (RCA4) are used for each ring in which variable torsions are used. During the run, random variations will be applied to 2-4 randomly selected dihedral angles from the TORS lists or molecular translations/rotations from the MOLS lists for each Monte Carlo step. If some other range of variable coordinates is desired, the range is set with the MCNV command.

If used within a BGIN/END loop, all structures in the input file will be read and the global minimum found will be listed to the output file at the end of every MC iteration set for each input structure. The ultimate global minimum would be found by examining the energies of each of the output structures.

Unlike MCMM, MCSM requires no MINI command. It must appear after a READ command.

In general, Monte Carlo searches should use CHIG commands to maintain all chiral centers and TORC commands to hold double bond geometries constant. If chirality or double-bond geometry is lost in any step, then following steps will be wrong if the resulting geometry is used for subsequent steps.

MCNV — Monte Carlo Number of Variables

This command resets the degrees of freedom (number of torsion angles varied plus number of molecules moved in space) altered in a single search step. If this command is not used, the value is taken from arg3 of the MCMM command. This command overrides the MCMM arg3.

For MCMM on single unsymmetrical molecules, we find it best to specify the range as 1 to N, where N is the number of variable dihedral angles, when TORS commands are being used. It is best to use a range of values, rather than a single value. When MOLS is being used, N should be incremented by one. When ZMAT is being used, N should take account of all degrees of freedom.

Used with MCMM, MCSD, MCLO, IMPS, and SPMC commands.

In the context of MC or MC(JBW) simulations this command sets the number of degrees of freedom to be changed at each MC step or randomized at each MC(JBW) step. A degree of freedom is either a bond length, bond angle or torsion or a molecular translation or rotation along or about a single axis. This number should be set in such a way to provide a compromise between acceptance rate and conformational interconversions. When args 1 and 2 of this command differ, MCNV defines a range for the number of degrees of freedom to be changed at each step. When the two arguments are identical, MCNV defines an exact number for the degrees of freedom to be changed at each step. In both cases, the initial values will be modified during the run by the adaptive mechanism unless debug 103 is defined.

For MC, MC(JBW) or MCSD simulations, a number of degrees of freedom must be defined and consequently, the MCNV command should be present; however, its two arguments can be 0, in which case the program will provide a default range, from 1 to maximum number of the degrees of freedom of the molecules. Such a range is probably not efficient. For MC(JBW)/SD, no randomization in the JBW part is needed and all the randomization can effectively be done by the SD part of the simulation. Doing so will increase the number of conformational interconversions. In order to achieve that, the MCNV command should be omitted and arg 2 of the MCSD command should be set to a negative number.

SEED — random number generator SEED

This command sets a starting value for the random number generator. Used to start the Monte Carlo random number generator at a different point so that repeated Monte Carlo or molecular dynamics runs give different results. The starting value should not exceed 700 000; if it does it will be divided by 2 until it is less than 700 000.

TORS — variable TORSion selection

Each TORS command specifies up to two torsions, using the numbers of the two central atoms. These will be used as variable dihedral angles by the MCMM, SPMC or MCSM commands. The actual number of torsions which will be varied during a single Monte Carlo step depends on the search method, but the number varied will be taken from the list specified in TORS commands. A given random torsional variation will be plus or minus a random number selected from the range extending from the arg5 to the arg6 specification.

Variable torsions within rings require the ring-closure commands RCA4 in addition to TORS commands. It is advisable to specify at least two variable torsions within each ring containing RCA4 ring closures. The atoms of a ring closures (args 2 and 3 in RCA4) must not be listed as variable torsions.

The minimum and maximum angular increment (arg5 and arg6) refer to the torsions given in arg1-4. It is possible to use a different angular increment for each torsion by using only arg1, arg2, arg5, and arg6 and a different TORS command for each torsion. For global searching, arg5 and arg6 of 0.0° and 180.0° are appropriate values. If you wish to focus the search on conformations having only small angular variations from the starting conformation, a value of 30.0° for arg6 could be used.

If you are searching multicyclic ring systems, you should include CHIG commands for any substituted atoms at the ring closure atoms (arg2 and arg3 in RCA4) to assure maintenance of stereochemistry.

When doing substructure MC searching, always order the pairs of atoms defining torsions such that the second atom of each pair is not connected to any fixed atoms (FXAT) except via the first atom (in the torsional movements, the chain connected to the second atom is the one that will actually be moved). If both ends are anchored by FXAT commands, then a ring closure (RCA4) command will be necessary. TORS and MOLS commands can be used together.

TRES — Torsional RESolution

This command may be used in the SPMC systematic pseudo Monte Carlo searches to alter the initial resolution of the search around a particular torsion angle. This command must come after the TORS command.

MOLS — variable MOLeculeS selection

This command selects molecules to be independently rotated and/or translated during a conformational search. It is used for configurational/conformational searches of complexes. In particular, given a docked bimolecular complex, MOLS can be used to translate and rotate the smaller molecule within the binding site of the larger one in order to explore possible binding geometries. MOLS and TORS commands can be used together; this allows the internal geometry of the separate molecules to be explored together with the relative orientation. If there are N molecules in the system, it suffices to specify N−1 of them in MOLS commands to ensure that all are adjusted.

MOLS commands must come after TORS commands.

During a search, random molecules are selected for motion. For each molecule, rotation about all three axes and translation along all three axes are performed by amounts selected randomly from within the ranges specified in arg5-6 and arg7-8. MOLS has the same relationship to LIGB as TORS has to RCA4: the bonds specified in a LIGB command are broken before the MOLS-specified molecular motion is carried out, then the LIGB bonds are remade prior to minimization of the resulting structure.

Note: During a long enough search, if pair-list cutoffs are in effect, as is normal (see EXNB), one molecule may eventually wander far enough away from another that no nonbonded energies between them exist anymore. In this situation, further searching just explores random spatial dispositions of this pair with no energetic contribution from their mutual interaction. Unless already-found binding conformations are lower in energy by at least the DEMX-specified energy, this will lead to essentially an infinite and fruitless search of conformational space. To avoid this, use FXDI to constrain the distance between atom pairs spanning pairs of molecules to some maximum distance. This maximum distance should be specified as the half-width of an FXDI potential. In practice, we do this only if we encounter difficulties without doing so.

LIGB — LIGand Bonds

Used chiefly for configurational searches of inorganic complexes in conjunction with the VDWB command.

This command defines bonds to be broken, creating molecular fragments which will be moved independently during the search. A MOLS command must be present for each fragment to be moved. For a bidentate ligand, there will be two LIGB bonds specified for the single MOLS command that moves the ligand; for a tridentate ligand, there will be three, and so on.

This procedure allows an MCMM search to find, for example, both mer and fac isomers of an octahedral complex with stoichiometry MA₃B₃. LIGB can also be used to extend a conformational search to a configurational search; for example, by specifying bonds to chiral carbons as LIGB bonds, the R and S configurations about these carbons will be explored.

RCA4 — Ring Closure Atoms

This command directs the program to temporarily break a bond to sever a ring for the purpose of Monte Carlo torsion angle searching. The first 4 arguments are four atoms within the ring which comprise a ring-closure torsion angle. One RCA4 command is necessary for each ring with dihedral angles specified in a TORS command. Used with MCMM, SPMC, IMPS, and MCSM.

With MCMM and MCSM, and for small-medium rings, arg5 and arg6 should be approximately 0.5 and 2.5. For large rings, arg5 and arg6 should be ca 0.1 and 5.0.

With SPMC and for small-medium rings, arg5 and arg6 should be approximately 1.0 and 2.0. For larger rings, arg5 and arg6 should be ca 0.5 and 3.0-4.0. The exact choice is not very important but has a minor effect on search efficiency.

It is forbidden that an arg2 or arg3 atom be used in more than one RCA4 command (common ring closure atoms are not allowed). The closure angle arguments (arg7 and arg8) are optional and refer to both closure angles 1-2-3 and 2-3-4. Our tests so far indicate that there is little reason to use arg7 and arg8 closure angle constraints.

DISC — DIStance Check

This command causes the distance between atoms given in arg1 and arg2 to be monitored and structures which have such distances outside the allowable range to be eliminated from the output file. Used only with MULT, MCSM, and MCMM.

TORC — TORsional Check

This command causes the absolute value of the torsion angle defined by arg1-4 to be monitored and structures which have such torsions outside the allowable range to be eliminated from the output file. Used only with MULT, MCSM, and MCMM. The test is applied after minimization in each case, and acts as a filter, particularly during conformational searches.

This command is intended to reject unwanted cis or trans isomers created during a conformation search. Appropriate values for rejecting all trans isomers of a cis torsion (original torsion angle ~0) are arg5=0 and arg6=90. To reject all cis isomers of a trans torsion (original torsion angle ~180), use arg5=90, and arg6=180.

Note: Torsion checks do not limit the range of conformers searched or constrain the torsion during minimization. See FXTA for details on applying a torsional constraint during minimization.

MCMF — Monte Carlo Maximum constraint Failures

This command allows you to adjust the maximum number of constraint failures, e.g. ring closures (RCA4) or constrained torsion (TORC), that are allowed before accepting a faulty structure. A limit of some kind is appropriate since the user could supply constraints that make valid structure generation impossible. Thus the program tries a given starting geometry and a given number of varying torsions (or other coordinates) repeatedly until arg1 failures (or the default of 10000) have occurred. Then, the program allows a different number of varying torsions and makes arg1 new attempts to create a valid structure. This process is repeated arg2 times before a faulty structure is accepted. Thus when constraint tests have failed arg1 times, the program allows arg2 failures (of arg1 tries each) before accepting the structure anyway. Such faulty structures often fail to minimize properly.

SMPL — monte carlo SaMPLing

This command allows one to write sample structures to disk during a Monte Carlo Single Minimum (MCSM) run.

SRNO— SeRial Number Ordering