User-defined restraint forms

To create a new restraint form, derive a new class from the base forms.RestraintForm. You should then override the following functions: __init__, eval, vmin, rvmin, min_mean, vheavy, rvheavy, heavy_mean, and get_range. Note that presently you can only derive from this base class, not from MODELLER built-in forms.

Restraint forms can act on one or more features (each of which has an accompanying integer modality, which you can use for any purpose), and can take any number of floating-point parameters as input. The features and parameters are stored in self._features and self._parameters respectively, but for convenience the base constructor RestraintForm.__init__ can set initial values for these.

The eval function is called from MODELLER with the current feature values, their types and modalities, and the parameter vector. You should return the objective function contribution and, if requested, the derivatives with respect to each feature. The feature types are required by the deltaf function, which returns the difference between the current feature value and the mean (a simple subtraction is not sufficient, as some feature types are periodic). Note that you must use the passed parameter vector, as the class is not persistent, and as such the self._parameters variable (or any other object variable you may have set) is not available to this function.

The get_range function is used to define the feature range over which the form is clearly non-linear. It is simply passed a similar set of parameters to eval, and should return a 2-element tuple containing the minimum and maximum feature values. It is only necessary to define this function if the form acts on only a single feature and you want to be able to convert it to a cubic spline using Restraints.spline().

The other functions are used to return the minimal and heavy restraint violations (both absolute and relative; see Section 5.3.1) and the means. The heavy and minimal means correspond to the global and local minima.

Example: examples/python/user_form.py

from modeller import *
from modeller.scripts import complete_pdb

env = Environ()

env.io.atom_files_directory = ['../atom_files']
log.verbose()
env.libs.topology.read(file='$(LIB)/top_heav.lib')
env.libs.parameters.read(file='$(LIB)/par.lib')

class MyGauss(forms.RestraintForm):
    """An implementation of Modeller's harmonic/Gaussian restraint (type 3)
       in pure Python"""

    rt = 0.5900991    # RT at 297.15K, in kcal/mol

    def __init__(self, group, feature, mean, stdev):
        forms.RestraintForm.__init__(self, group, feature, 0, (mean, stdev))

    def eval(self, feats, iftyp, modal, param, deriv):
        (mean, stdev) = param
        delt = self.deltaf(feats[0], mean, iftyp[0])
        val = self.rt * 0.5 * delt**2  / stdev**2
        if deriv:
            fderv = self.rt * delt / stdev**2
            return val, [fderv]
        else:
            return val

    def vmin(self, feats, iftyp, modal, param):
        (mean, stdev) = param
        return self.deltaf(feats[0], mean, iftyp[0])

    def rvmin(self, feats, iftyp, modal, param):
        (mean, stdev) = param
        return self.deltaf(feats[0], mean, iftyp[0]) / stdev

    def min_mean(self, feats, iftyp, modal, param):
        (mean, stdev) = param
        return [mean]

    def get_range(self, iftyp, modal, param, spline_range):
        (mean, stdev) = param
        return (mean - stdev * spline_range, mean + stdev * spline_range)

    # There is only one minimum, so the 'heavy' mean is the same as the 'min'
    vheavy = vmin
    rvheavy = rvmin
    heavy_mean = min_mean

mdl = complete_pdb(env, "1fdn")
sel = Selection(mdl)
rsr = mdl.restraints
at = mdl.atoms
rsr.add(MyGauss(group=physical.bond,
                feature=features.Distance(at['CB:1:A'], at['CA:1:A']),
                mean=1.5380, stdev=0.0364))
sel.energy()

# Restraints using user-defined forms can be converted to splines for speed.
# Only one-dimensional forms that define the get_range() method can be splined.
rsr.spline(MyGauss, features.Distance, physical.bond, spline_dx=0.05)
sel.energy()