Next: Overview of Directed Searching, Previous: Backbone and Chain Closure Overview, Up: Overview of Conformational Search

Given a set of backbone conformations, it remains to generate a set of side chain atom positions for each of the backbone conformations. Before we explore the problems inherent in side chain generation, we describe the side chain atom placement.

As with the backbone atom placement, the side chain atoms are
positioned based on free torsion angles. The side chain torsions are
processed from the backbone out as each succeeding atom requires the
position of the previous atom for its placement. The sampling interval
of each torsion (the option `SGRID`) can be either some fixed number
of degrees or the period of the torsion energy. When the latter is used,
the sidechain torsions will be at minima in the torsion angle potential
involving the free atom and its antecedents.
It is also possible to
modify the sampling to avoid van der Waals contacts (`VAVOID`
option). It is common for one free torsion to generate the position of
more than one atom because of side chain branching, non-rotatable bonds,
and rings. For example, although tryptophan has 11 side chain atoms to
be placed, it has only two free torsion angles. Also, certain torsions
have symmetry so we can reduce the sampling necessary. Finally, a search
of the surrounding space is made for any constructed atom to see if
there are any close contacts with a repulsive energy greater than
`MAXEVDW`, and if so, that structure is eliminated.

Although this data structure was designed for amino acids, it can be applied to an arbitrary molecule. The only prerequisite is the presence of a few known atomic position upon which the remaining atoms can be constructed.

The information needed for side chain construction is stored in a side chain topology file, see Sidechain Topology.

Given these specifications for generating side chain atomic positions, we need to introduce a protocol that generates only a limited number of conformers. The procedure analogous to the backbone generation procedure would result in a series of nested iterations over each chi torsion angle with the number of levels being equal to the sum of the free torsions in all the side chains of the peptide segment. The large number of free torsions in the side chains and the absence of a connectivity constraint, such as exists for the backbone, result in an enormous number of possible sidechain conformations. Consequently, such a direct approach is not feasible except in limited cases.

However, the situation is not that bleak. First, the backbone construction process provides the position of CB which gives a strong bias to the side chain orientation. Thus, an acceptable course of action is the generation of only one sidechain conformation for each backbone conformation. We must strive to make this one conformation the lowest energy possible for the given backbone. Second, because the side chains close together in sequence frequently are not close together in space, and therefore, do not interact strongly, it is a reasonable approximation to treat the side chains quasi-independently. Instead of finding all combinations of side chain atomic positions, we can handle the side chains sequentially so the time required for side chain placement increases linearly, rather than exponentially, with the number of residues.

In order not to limit the options for using the program, six
possible methods for generating side chain positions have been
implemented. These are specified using the `SIDEOPT` option in the
sidechain degree of freedom. All of these methods discard conformations
which have any repulsive contacts exceeding `MAXEVDW` in van der
Waals energy. The first two methods described, `ALL` and
`FIRST`, assume no quasi-independence of the sidechains whereas the
others do.

The first method, `ALL`, generates all possible
conformations by a series of nested iterations over every sidechain as
described above. The second method, `FIRST`, uses the same algorithm
as `ALL` except that all the iterations terminate when the first
conformation for all the sidechains has been found. This method is
useful for determining if a backbone conformation will accommodate the
sidechains when details about the sidechain energetics are not
required.

The next three methods all depend on a function which evaluates the side chain positions as they are generated so that the best ones can be selected. "Best" is defined as the conformation whose evaluation function is numerically smallest. Two evaluation functions are currently provided, one based on positional deviations, and one based on the CHARMM energy function. The evaluation function based on positional deviations is present for testing CONGEN as it provides a means for determining the limit of CONGEN's ability to generate a known structure. If coordinates are present for the peptide gap, this evaluation function will determine the RMS shift between a generated side chain conformation and the initial coordinates. The second evaluation function computes the CHARMM energy of the sidechain atoms omitting the bond and bond angle energies because the generation procedure does not vary either of these two terms. At present, either the r dependent dielectric or the constant dielectric for the electrostatic energy is used. The other electrostatic calculations, see Non-bonded Interactions, are not available.

The `INDEPENDENT` method assumes
that the side chains in the peptide chain being generated do not
interact with one another. The atoms of each side chain are placed
independently, with those of the other side chains in the peptide being
ignored; interactions with all other atoms in the system are included.
The conformation which has the lowest value for the evaluation function
is selected for each side chain. When the RMS evaluation function is
used, this method gives the optimum conformation, though it may be
sterically inappropriate. Thus, it cannot be used when the energy is the
evaluation function unless the possibility of large repulsive van der
Waals is not important.

The method, `COMBINATION`, begins by generating a small
number of the best side chain conformations for each side chain
independently, as above. Then, these side chain conformations are
assembled in all possible combinations, and those combinations which do
not have bad van der Waals contacts are accepted. The number of
conformations saved for each side chain must be small to avoid a
combinatorial explosion.

The `ITERATIVE` method starts with an energetically
acceptable side chain conformation for all the side chains. This
conformation is generated, if possible, using the `FIRST` method
(see above). Starting with this conformation, we regenerate all the
possible positions for the side chain atoms of the first residue, and
select the conformation with the lowest energy. We also save the value
of the evaluation function. This regeneration is done with all the other
side chain atoms present so we can account for their effect. The process
is repeated sequentially for the rest of sidechains in the gap. We then
return to the first residue and go through the process again until the
energies of the side chain atoms do not change or until the number of
passes reaches an iteration limit. This method has the virtue that only
one conformation is generated per backbone conformation, and it is an
energetically reasonable one. However, if there are significant
interactions between the sidechain atoms, the first part of the process
will bias the iterative process toward the initial side chain
arrangement selected, and we may miss the lowest energy side chain
conformation.

The `FIXED` method is used to construct sidechains in a fixed
conformation. When this method is specified, the program calculates the construction
bond lengths, angles, and torsions for all atoms in the degree of freedom
from the starting coordinates, and will generate just one sample using
those values. If there are van der Waals overlaps, then no conformation will
be generated.

With any of the methods described above, the `CONGEN`
command can apply any of the minimization algorithms to the generated
conformations before they are written out for further analysis.
Minimization provides an ability to reduce the small van der Waals
repulsions that are inevitable with coarse torsion grids used.

See Sidechain Degree of Freedom, for more information on this degree of freedom.