First version of coordinate generatign processor

vermouth/processors/__init__.py

@@ -44,3 +44,4 @@
 from .sort_molecule_atoms import SortMoleculeAtoms
 from .merge_all_molecules import MergeAllMolecules
 from .annotate_mut_mod import AnnotateMutMod
+from .generate_coordinates import GenerateCoordinates

vermouth/processors/generate_coordinates.py

@@ -0,0 +1,368 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# Copyright 2018 University of Groningen
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Provides a processor for building missing coordinates.
+"""
+
+import numpy as np
+import networkx as nx
+
+from .processor import Processor
+from .. import selectors
+
+
+def mindist(A, B):
+    '''
+    Determine the smallest distance between two point sets (np.ndarray) A and B
+    '''
+    return ((A[:, None] - B[None])**2).sum(axis=2).min(axis=1)
+
+
+def get_atoms_missing_coords(molecule):
+    '''
+    Determine particles without coordinates in molecule.
+    '''
+    return [
+        ix for ix in molecule
+        if not selectors.selector_has_position(molecule.nodes[ix])
+    ]
+
+
+def get_anchors(molecule, indices):
+    '''
+    Determine particles with known coordinates connected to
+    particles with given indices.
+    '''
+    return {
+        a for ix in indices for a in molecule[ix].keys()
+        if selectors.selector_has_position(molecule.nodes[a])
+    }
+
+
+def align_z(v):
+    '''
+    Generate rotation matrix aligning z-axis onto vector v (and vice-versa).
+    '''
+    # This routine is based on the notion that for any two
+    # normalized vectors, the alignment is a 180 degree
+    # rotation around the resultant vector.
+
+    w = v / (v ** 2).sum() ** 0.5
+
+    if w[2] <= 1e-8 - 1:
+        return -np.eye(3) # Mind the inversion ...
+
+    w[2] += 1
+
+    return (2 / (w**2).sum()) * w * w[:, None] - np.eye(3)
+
+
+def conicring(points=24, distance=0.125, angle=109.4, axis=None):
+    '''
+    Create a ring of positions corresponding to a cone with given angle.
+    An angle of 109.4 is suitable for ideal SP3 substituents.
+    '''
+    p = 2 * np.pi * np.arange(points) / points
+    r = distance * np.sin(angle * np.pi / 360)
+    z = distance * np.cos(angle * np.pi / 360)
+    X = np.stack([r * np.cos(p), r * np.sin(p), z * np.ones(points)], axis=1)
+    if axis is None:
+        return X
+    return X @ align_z(axis)
+
+
+def double_cone(points=24, distance=0.125, angle=109.4, axis=None):
+    '''Create positions for two consecutive (SP3) bonded particles'''
+    P = conicring(points, distance, angle, axis)
+    S = np.array([P @ align_z(x) + x for x in P])
+    return P, S
+
+
+def triple_cone(points=24, distance=0.125, angle=109.4):
+    '''Create possible positions for three consecutive (SP3) bonded particles'''
+    P, S = double_cone(points, distance, angle)
+    T = np.array([[[P @ align_z(u - x) + u] for u in U] for x, U in zip(P, S)])
+    return P, S, T
+
+
+def _out(molecule, anchor, base, distance=0.125):
+    '''
+    Generate coordinates for atom bonded to anchor
+    using resultant vector of known substituents.
+
+    This approach works for placement of single unknown
+    substituents on sp1, sp2 or sp3 atoms. It should also
+    work for bipyramidal and other configurations.
+
+     b1
+       \
+    b2--a-->X
+       /
+     b3
+
+    X := position to determine
+    a := position of anchoring particle
+    B = [b1, b2, ...] := positions of base particles
+
+    Parameters
+    ----------
+    molecule: vermouth.molecule.Molecule
+        A molecule with missing atoms
+    anchor: integer
+        Index to anchor node: a node with positions connected to a node without
+    base: list of integers
+        Indices to particles with positions connected to anchor
+    distance: float
+        Distance of target position from anchor
+
+    Returns
+    -------
+        position
+    '''
+    a = molecule.nodes[anchor]['position']
+    B = [ molecule.nodes[ix]['position'] for ix in base ]
+    B = B - a
+    u = -B.sum(axis=0)
+    u *= distance / ((u ** 2).sum() ** 0.5)
+    return a + u
+
+
+def _chiral(molecule, anchor, base, distance=(0.125, 0.125)):
+    '''
+    Generate coordinates for atoms bonded to chiral center
+    based on two known substituents.
+
+    This approach uses the positions of two substituents to
+    generate coordinates for one or two atoms connected to
+    a chiral center. The two atoms are indicated as `up` and
+    `down`, referring to the direction with respect to the
+    cross-product of the two substituents. If these control
+    atoms are not given explicitly in order, then the control
+    atoms are inferred from the connections of the anchor with
+    known coordinates. In that case, the specific chirality
+    of the center (R or S) cannot be controlled.
+
+    NOTE: This function should probably be able to determine
+          the weights of the substituents to allow placement
+          based on R or S chirality.
+
+    Parameters
+    ----------
+    molecule: vermouth.molecule.Molecule
+        A molecule with missing atoms
+    anchor: integer
+        Index to anchor node: a node with positions connected to a node without
+    base: list of integers
+        Indices to particles with positions connected to anchor
+    distance: tuple of two floats
+        Distances of target positions from anchor
+
+    Returns
+    -------
+        tuple of positions (up, down)
+    '''
+    a = molecule.nodes[anchor]['position']
+    B = [ molecule.nodes[ix]['position'] for ix in base ]
+    B = B - a
+    # Negative resultant vector
+    u = -0.5 * (B[0] + B[1])
+    v = np.cross(B[0], B[1])
+    # Set length of v to length to half distance between subs
+    v = 0.5 * ((B[0] - B[1])**2).sum()**0.5 * v / (v ** 2).sum()**0.5
+    uv = u + v
+    # Normalization of vector sum
+    n = 1 / (uv**2).sum()**0.5
+
+    rup, rdown = distance
+
+    pup = a + rup * n * uv
+    pdown = a + rdown * n * (u - v)
+
+    return pup, pdown
+
+
+class Segment:
+    '''Class for subgraphs of (missing) atoms with anchors'''
+    def __init__(self, molecule, limbs):
+        self.molecule = molecule
+        self.anchors = limbs[0][0]
+        self.bases = limbs[0][1]
+        self.limbs = [ s[2] for s in limbs ]
+
+    def __str__(self):
+        return ':'.join([str(self.anchors), str(self.bases), str(self.limbs)])
+
+    def extrude(self, distance=0.125, coords=None, contact=0.5):
+        '''
+        Generate positions for particles without, connected to anchors.
+        '''
+        mol = self.molecule
+
+        for i, (a, b) in enumerate(zip(self.anchors, self.bases)):
+            target = [ k for k in mol.neighbors(a) if k not in b ]
+            valence = len(target) + len(b)
+
+            if valence > 2 and len(b) == valence - 1:
+                # Trivial chiral/flat/bipyramidal/octahedral/...
+                mol.nodes[target[0]]['position'] = _out(mol, anchor=a, base=b)
+                for k in self.limbs:
+                    k.discard(target[0])
+                # Simply update this segment and return it
+                self.bases[i] = [a]
+                self.anchors[i] = target[0]
+                return [self]
+
+            if valence == 4 and len(b) == 2:
+                # Trivial chiral - except for the actual chirality
+                # a simple 'up' or 'down' flag would be helpful
+                # amino acid side chain is 'up'
+                up, down = _chiral(mol, anchor=a, base=b)
+                mol.nodes[target[0]]['position'] = up
+                mol.nodes[target[1]]['position'] = down
+                # Simply update this segment and return it
+                for k in self.limbs:
+                    k.discard(target[0])
+                self.bases[i] = [a]
+                self.anchors[i] = target[0]
+                return [self]
+
+            if valence == 1:
+                # Generate a sphere of points
+                # Check for overlaps
+                # Take a pick
+                # Should be done last anyway
+                # Best collected and stored
+                continue
+
+            # So a valence of 2 (missing 1), 3 (missing 2) or 4 (missing 3)
+            # The cases 2(1), 3(2), 4(3) can be solved by sampling possible
+            # candidate points from a conical set.
+
+            if len(b) == 1 and all(len(s) < 100 for s in self.limbs):
+                ax = mol.nodes[a]['position']
+                bx = mol.nodes[b[0]]['position']
+                P = conicring(axis=ax-bx) + ax
+                if coords is not None:
+                    P[mindist(P, coords) < contact**2] = np.nan
+                P = P.reshape((len(self.limbs), -1, 3))
+                for px in P.transpose((1,0,2)):
+                    if px.max() == np.nan:
+                        continue
+                    # Split the limbs to new segments and return those
+                    #out = []
+                    for t, pi in zip(target, px):
+                        self.molecule.nodes[t]['position'] = pi
+                    #    for k in self.limbs:
+                    #        if len(k) > 1 and t in k:
+                    #            k.discard(t)
+                    #            for g in nx.connected_components(self.molecule.subgraph(k)):
+                    #                out.append(
+                    #                    Segment(
+                    #                        self.molecule,
+                    #                        [(t, [[a]], g)],
+                    #                        self.coords,
+                    #                        self.contact
+                    #                    )
+                    #                )
+                    #  This commented-out stuff may be useful for speeding up
+                    #  coordinate generation, because it avoids needing lookup
+                    #  of missing coordinates in subsequent cycles.
+                    #return out
+                    return
+                break
+        return None
+
+
+class GenerateCoordinates(Processor):
+    """
+    A processor for generating coordinates for atoms without.
+    """
+    def __init__(self, contact=0.5):
+        super().__init__()
+        self.contact = contact
+        self.coordinates = None
+
+    def run_system(self, system):
+        """
+        Process `system`.
+
+        Parameters
+        ----------
+        system: vermouth.system.System
+            The system to process. Is modified in-place.
+        """
+
+        self.coordinates = np.stack(
+            [
+                a.get('position', (np.nan, np.nan, np.nan))
+                for m in system.molecules
+                for i, a in m.atoms
+            ]
+        )
+
+        for molecule in system.molecules:
+            self.run_molecule(molecule)
+
+    def run_molecule(self, molecule):
+        """
+        Process a single molecule. Must be implemented by subclasses.
+
+        Parameters
+        ----------
+        molecule: vermouth.molecule.Molecule
+            The molecule to process.
+
+        Returns
+        -------
+        vermouth.molecule.Molecule
+            The provided molecule with complete coordinates
+        """
+
+        # Missing atoms - those without 'positions'
+        missing = get_atoms_missing_coords(molecule)
+
+        prev = len(molecule)
+        while prev - len(missing) > 0:
+            # Limbs: subgraphs of missing atoms with anchors and bases
+            limbs = []
+            for g in nx.connected_components(molecule.subgraph(missing)):
+                anchors = list(get_anchors(molecule, g))
+                bases = [ list(get_anchors(molecule, [a])) for a in anchors ]
+                limbs.append((anchors, bases, g))
+            limbs.sort()
+
+            # Segment: subgraphs of limbs sharing anchors and bases
+            segments = []
+            prev = None
+            for lim in limbs:
+                if lim[0] != prev:
+                    segments.append([])
+                    prev = lim[0]
+                segments[-1].append(lim)
+            segments = [ Segment(molecule, limbs) for limbs in segments ]
+
+            # Extrude segments one layer at a time
+            while segments:
+                s = segments.pop(0)
+                r = s.extrude(coords=self.coordinates, contact=self.contact)
+                if r is not None:
+                    segments.extend(r)
+
+            prev = len(missing)
+            missing = get_atoms_missing_coords(molecule)
+            #print(len(missing))
+
+        return molecule