pub_soft/rdkit
3
0
Fork 0
mirror of https://github.com/rdkit/rdkit.git synced 2026-06-04 21:54:27 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
a59997b7e4fe5098dfb3bd7fcd95b3dc2b8c8806
rdkit/Code/GraphMol/DistGeomHelpers/Embedder.cpp
Greg Landrum a59997b7e4 Some refactoring of the distance geometry code (#2153) 
* move _embedPoints over to use EmbedArgs structure

* move _embedWithExpTorsions over

* a few pointer cleanups

* step by step

* more

* wrap the ETKDG paramaters into a struct

* fix a typo

* additional cleanup

* more

* simplify the embedargs struct

* pass embedParams instead of having them in embedargs

* a bit more cleanup

* response to code review
2019-01-02 06:06:38 +00:00

993 lines
38 KiB
C++
Raw Blame History

//
// Copyright (C) 2004-2018 Greg Landrum and Rational Discovery LLC
//
// @@ All Rights Reserved @@
// This file is part of the RDKit.
// The contents are covered by the terms of the BSD license
// which is included in the file license.txt, found at the root
// of the RDKit source tree.
//
// define DEBUG_EMBEDDING 1
#include "Embedder.h"
#include <DistGeom/BoundsMatrix.h>
#include <DistGeom/DistGeomUtils.h>
#include <DistGeom/TriangleSmooth.h>
#include <DistGeom/ChiralViolationContrib.h>
#include "BoundsMatrixBuilder.h"
#include <ForceField/ForceField.h>
#include <GraphMol/ROMol.h>
#include <GraphMol/Atom.h>
#include <GraphMol/AtomIterators.h>
#include <GraphMol/RingInfo.h>
#include <GraphMol/Conformer.h>
#include <RDGeneral/types.h>
#include <RDGeneral/RDLog.h>
#include <RDGeneral/Exceptions.h>
#include <Geometry/Transform3D.h>
#include <Numerics/Alignment/AlignPoints.h>
#include <DistGeom/ChiralSet.h>
#include <GraphMol/MolOps.h>
#include <GraphMol/ForceFieldHelpers/CrystalFF/TorsionPreferences.h>
#include <boost/dynamic_bitset.hpp>
#include <iomanip>
#include <RDGeneral/RDThreads.h>
#ifdef RDK_THREADSAFE_SSS
#include <future>
#endif
//#define DEBUG_EMBEDDING 1
namespace {
const double ERROR_TOL = 0.00001;
// these tolerances, all to detect and filter out bogus conformations, are a
// delicate balance between sensitive enough to detect obviously bad
// conformations but not so sensitive that a bunch of ok conformations get
// filtered out, which slows down the whole conformation generation process
const double MAX_MINIMIZED_E_PER_ATOM = 0.05;
const double MAX_MINIMIZED_E_CONTRIB = 0.20;
const double MIN_TETRAHEDRAL_CHIRAL_VOL = 0.50;
const double TETRAHEDRAL_CENTERINVOLUME_TOL = 0.30;
} // namespace
namespace RDKit {
namespace DGeomHelpers {
//! Parameters corresponding to Sereina Riniker's KDG approach
const EmbedParameters KDG(0, // maxIterations
1, // numThreads
-1, // randomSeed
true, // clearConfs
false, // useRandomCoords
2.0, // boxSizeMult
true, // randNegEig
1, // numZeroFail
NULL, // coordMap
1e-3, // optimizerForceTol
false, // ignoreSmoothingFailures
true, // enforceChirality
false, // useExpTorsionAnglePrefs
true, // useBasicKnowledge
false, // verbose
5.0, // basinThresh
-1.0, // pruneRmsThresh
true, // onlyHeavyAtomsForRMS
1 // ETversion
);
//! Parameters corresponding to Sereina Riniker's ETDG approach
const EmbedParameters ETDG(0, // maxIterations
1, // numThreads
-1, // randomSeed
true, // clearConfs
false, // useRandomCoords
2.0, // boxSizeMult
true, // randNegEig
1, // numZeroFail
NULL, // coordMap
1e-3, // optimizerForceTol
false, // ignoreSmoothingFailures
false, // enforceChirality
true, // useExpTorsionAnglePrefs
false, // useBasicKnowledge
false, // verbose
5.0, // basinThresh
-1.0, // pruneRmsThresh
true, // onlyHeavyAtomsForRMS
1 // ETversion
);
//! Parameters corresponding to Sereina Riniker's ETKDG approach
const EmbedParameters ETKDG(0, // maxIterations
1, // numThreads
-1, // randomSeed
true, // clearConfs
false, // useRandomCoords
2.0, // boxSizeMult
true, // randNegEig
1, // numZeroFail
NULL, // coordMap
1e-3, // optimizerForceTol
false, // ignoreSmoothingFailures
true, // enforceChirality
true, // useExpTorsionAnglePrefs
true, // useBasicKnowledge
false, // verbose
5.0, // basinThresh
-1.0, // pruneRmsThresh
true, // onlyHeavyAtomsForRMS
1 // ETversion
);
//! Parameters corresponding to Sereina Riniker's ETKDG approach - version 2
const EmbedParameters ETKDGv2(0, // maxIterations
1, // numThreads
-1, // randomSeed
true, // clearConfs
false, // useRandomCoords
2.0, // boxSizeMult
true, // randNegEig
1, // numZeroFail
NULL, // coordMap
1e-3, // optimizerForceTol
false, // ignoreSmoothingFailures
true, // enforceChirality
true, // useExpTorsionAnglePrefs
true, // useBasicKnowledge
false, // verbose
5.0, // basinThresh
-1.0, // pruneRmsThresh
true, // onlyHeavyAtomsForRMS
2 // ETversion
);
namespace detail {
struct EmbedArgs {
boost::dynamic_bitset<> *confsOk;
bool fourD;
INT_VECT *fragMapping;
std::vector<Conformer *> *confs;
unsigned int fragIdx;
DistGeom::BoundsMatPtr mmat;
DistGeom::VECT_CHIRALSET const *chiralCenters;
DistGeom::VECT_CHIRALSET const *tetrahedralCarbons;
ForceFields::CrystalFF::CrystalFFDetails *etkdgDetails;
};
} // namespace detail
bool _volumeTest(const DistGeom::ChiralSetPtr &chiralSet,
const RDGeom::PointPtrVect &positions, bool verbose = false) {
RDGeom::Point3D p0((*positions[chiralSet->d_idx0])[0],
(*positions[chiralSet->d_idx0])[1],
(*positions[chiralSet->d_idx0])[2]);
RDGeom::Point3D p1((*positions[chiralSet->d_idx1])[0],
(*positions[chiralSet->d_idx1])[1],
(*positions[chiralSet->d_idx1])[2]);
RDGeom::Point3D p2((*positions[chiralSet->d_idx2])[0],
(*positions[chiralSet->d_idx2])[1],
(*positions[chiralSet->d_idx2])[2]);
RDGeom::Point3D p3((*positions[chiralSet->d_idx3])[0],
(*positions[chiralSet->d_idx3])[1],
(*positions[chiralSet->d_idx3])[2]);
RDGeom::Point3D p4((*positions[chiralSet->d_idx4])[0],
(*positions[chiralSet->d_idx4])[1],
(*positions[chiralSet->d_idx4])[2]);
// even if we are minimizing in higher dimension the chiral volume is
// calculated using only the first 3 dimensions
RDGeom::Point3D v1 = p0 - p1;
v1.normalize();
RDGeom::Point3D v2 = p0 - p2;
v2.normalize();
RDGeom::Point3D v3 = p0 - p3;
v3.normalize();
RDGeom::Point3D v4 = p0 - p4;
v4.normalize();
RDGeom::Point3D crossp = v1.crossProduct(v2);
double vol = crossp.dotProduct(v3);
if (verbose) std::cerr << " " << fabs(vol) << std::endl;
if (fabs(vol) < MIN_TETRAHEDRAL_CHIRAL_VOL) return false;
crossp = v1.crossProduct(v2);
vol = crossp.dotProduct(v4);
if (verbose) std::cerr << " " << fabs(vol) << std::endl;
if (fabs(vol) < MIN_TETRAHEDRAL_CHIRAL_VOL) return false;
crossp = v1.crossProduct(v3);
vol = crossp.dotProduct(v4);
if (verbose) std::cerr << " " << fabs(vol) << std::endl;
if (fabs(vol) < MIN_TETRAHEDRAL_CHIRAL_VOL) return false;
crossp = v2.crossProduct(v3);
vol = crossp.dotProduct(v4);
if (verbose) std::cerr << " " << fabs(vol) << std::endl;
if (fabs(vol) < MIN_TETRAHEDRAL_CHIRAL_VOL) return false;
return true;
}
bool _sameSide(const RDGeom::Point3D &v1, const RDGeom::Point3D &v2,
const RDGeom::Point3D &v3, const RDGeom::Point3D &v4,
const RDGeom::Point3D &p0, double tol = 0.1) {
RDGeom::Point3D normal = (v2 - v1).crossProduct(v3 - v1);
double d1 = normal.dotProduct(v4 - v1);
double d2 = normal.dotProduct(p0 - v1);
// std::cerr << " " << d1 << " - " << d2 << std::endl;
if (fabs(d1) < tol || fabs(d2) < tol) return false;
return !((d1 < 0.) ^ (d2 < 0.));
}
bool _centerInVolume(unsigned int idx0, unsigned int idx1, unsigned int idx2,
unsigned int idx3, unsigned int idx4,
const RDGeom::PointPtrVect &positions, double tol,
bool verbose = false) {
RDGeom::Point3D p0((*positions[idx0])[0], (*positions[idx0])[1],
(*positions[idx0])[2]);
RDGeom::Point3D p1((*positions[idx1])[0], (*positions[idx1])[1],
(*positions[idx1])[2]);
RDGeom::Point3D p2((*positions[idx2])[0], (*positions[idx2])[1],
(*positions[idx2])[2]);
RDGeom::Point3D p3((*positions[idx3])[0], (*positions[idx3])[1],
(*positions[idx3])[2]);
RDGeom::Point3D p4((*positions[idx4])[0], (*positions[idx4])[1],
(*positions[idx4])[2]);
// RDGeom::Point3D centroid = (p1+p2+p3+p4)/4.;
if (verbose) {
std::cerr << _sameSide(p1, p2, p3, p4, p0, tol) << " "
<< _sameSide(p2, p3, p4, p1, p0, tol) << " "
<< _sameSide(p3, p4, p1, p2, p0, tol) << " "
<< _sameSide(p4, p1, p2, p3, p0, tol) << std::endl;
}
bool res = _sameSide(p1, p2, p3, p4, p0, tol) &&
_sameSide(p2, p3, p4, p1, p0, tol) &&
_sameSide(p3, p4, p1, p2, p0, tol) &&
_sameSide(p4, p1, p2, p3, p0, tol);
return res;
}
bool _centerInVolume(const DistGeom::ChiralSetPtr &chiralSet,
const RDGeom::PointPtrVect &positions, double tol = 0.1,
bool verbose = false) {
if (chiralSet->d_idx0 ==
chiralSet->d_idx4) { // this happens for three-coordinate centers
return true;
}
return _centerInVolume(chiralSet->d_idx0, chiralSet->d_idx1,
chiralSet->d_idx2, chiralSet->d_idx3,
chiralSet->d_idx4, positions, tol, verbose);
}
bool _boundsFulfilled(const std::vector<int> &atoms,
const DistGeom::BoundsMatrix &mmat,
const RDGeom::PointPtrVect &positions) {
// unsigned int N = mmat.numRows();
// std::cerr << N << " " << atoms.size() << std::endl;
// loop over all pair of atoms
for (unsigned int i = 0; i < atoms.size() - 1; ++i) {
for (unsigned int j = i + 1; j < atoms.size(); ++j) {
int a1 = atoms[i];
int a2 = atoms[j];
RDGeom::Point3D p0((*positions[a1])[0], (*positions[a1])[1],
(*positions[a1])[2]);
RDGeom::Point3D p1((*positions[a2])[0], (*positions[a2])[1],
(*positions[a2])[2]);
double d2 = (p0 - p1).length(); // distance
double lb = mmat.getLowerBound(a1, a2);
double ub = mmat.getUpperBound(a1, a2); // bounds
if (((d2 < lb) && (fabs(d2 - lb) > 0.1 * ub)) ||
((d2 > ub) && (fabs(d2 - ub) > 0.1 * ub))) {
#ifdef DEBUG_EMBEDDING
std::cerr << a1 << " " << a2 << ":" << d2 << " " << lb << " " << ub
<< " " << fabs(d2 - lb) << " " << fabs(d2 - ub) << std::endl;
#endif
return false;
}
}
}
return true;
}
namespace EmbeddingOps {
bool generateInitialCoords(RDGeom::PointPtrVect *positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams,
RDNumeric::DoubleSymmMatrix &distMat,
RDKit::double_source_type *rng) {
bool gotCoords = false;
if (!embedParams.useRandomCoords) {
double largestDistance =
DistGeom::pickRandomDistMat(*eargs.mmat, distMat, *rng);
RDUNUSED_PARAM(largestDistance);
gotCoords = DistGeom::computeInitialCoords(distMat, *positions, *rng,
embedParams.randNegEig,
embedParams.numZeroFail);
} else {
double boxSize;
if (embedParams.boxSizeMult > 0) {
boxSize = 5. * embedParams.boxSizeMult;
} else {
boxSize = -1 * embedParams.boxSizeMult;
}
gotCoords = DistGeom::computeRandomCoords(*positions, boxSize, *rng);
}
return gotCoords;
}
bool firstMinimization(RDGeom::PointPtrVect *positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams) {
bool gotCoords = true;
std::unique_ptr<ForceFields::ForceField> field(DistGeom::constructForceField(
*eargs.mmat, *positions, *eargs.chiralCenters, 1.0, 0.1, nullptr,
embedParams.basinThresh));
unsigned int nPasses = 0;
field->initialize();
if (field->calcEnergy() > ERROR_TOL) {
int needMore = 1;
while (needMore) {
needMore = field->minimize(400, embedParams.optimizerForceTol);
++nPasses;
}
}
std::vector<double> e_contribs;
double local_e = field->calcEnergy(&e_contribs);
#ifdef DEBUG_EMBEDDING
std::cerr << " Energy : " << local_e / positions.size() << " "
<< *(std::max_element(e_contribs.begin(), e_contribs.end()))
<< std::endl;
#endif
// check that neither the energy nor any of the contributions to it are
// too high (this is part of github #971)
if (local_e / positions->size() >= MAX_MINIMIZED_E_PER_ATOM ||
(e_contribs.size() &&
*(std::max_element(e_contribs.begin(), e_contribs.end())) >
MAX_MINIMIZED_E_CONTRIB)) {
#ifdef DEBUG_EMBEDDING
std::cerr << " Energy fail: " << local_e / positions->size() << " "
<< *(std::max_element(e_contribs.begin(), e_contribs.end()))
<< std::endl;
#endif
gotCoords = false;
}
return gotCoords;
}
bool checkTetrahedralCenters(const RDGeom::PointPtrVect *positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams) {
RDUNUSED_PARAM(embedParams);
// for each of the atoms in the "tetrahedralCarbons" list, make sure
// that there is a minimum volume around them and that they are inside
// that volume. (this is part of github #971)
for (const auto tetSet : *eargs.tetrahedralCarbons) {
// it could happen that the centroid is outside the volume defined
// by the other
// four points. That is also a fail.
if (!_volumeTest(tetSet, *positions) ||
!_centerInVolume(tetSet, *positions, TETRAHEDRAL_CENTERINVOLUME_TOL)) {
#ifdef DEBUG_EMBEDDING
std::cerr << " fail2! (" << tetSet->d_idx0 << ") iter: " << iter
<< " vol: " << _volumeTest(tetSet, *positions, true)
<< " center: "
<< _centerInVolume(tetSet, *positions,
TETRAHEDRAL_CENTERINVOLUME_TOL, true)
<< std::endl;
#endif
return false;
}
}
return true;
}
bool checkChiralCenters(const RDGeom::PointPtrVect *positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams) {
RDUNUSED_PARAM(embedParams);
// check the chiral volume:
for (const auto chiralSet : *eargs.chiralCenters) {
double vol = DistGeom::ChiralViolationContrib::calcChiralVolume(
chiralSet->d_idx1, chiralSet->d_idx2, chiralSet->d_idx3,
chiralSet->d_idx4, *positions);
double lb = chiralSet->getLowerVolumeBound();
double ub = chiralSet->getUpperVolumeBound();
if ((lb > 0 && vol < lb && (lb - vol) / lb > .2) ||
(ub < 0 && vol > ub && (vol - ub) / ub > .2)) {
#ifdef DEBUG_EMBEDDING
std::cerr << " fail! (" << chiralSet->d_idx0 << ") iter: " << iter << " "
<< vol << " " << lb << "-" << ub << std::endl;
#endif
return false;
}
}
return true;
}
bool minimizeFourthDimension(RDGeom::PointPtrVect *positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams) {
// now redo the minimization if we have a chiral center
// or have started from random coords. This
// time removing the chiral constraints and
// increasing the weight on the fourth dimension
std::unique_ptr<ForceFields::ForceField> field2(DistGeom::constructForceField(
*eargs.mmat, *positions, *eargs.chiralCenters, 0.2, 1.0, nullptr,
embedParams.basinThresh));
field2->initialize();
// std::cerr<<"FIELD2 E: "<<field2->calcEnergy()<<std::endl;
if (field2->calcEnergy() > ERROR_TOL) {
int needMore = 1;
int nPasses2 = 0;
while (needMore) {
needMore = field2->minimize(200, embedParams.optimizerForceTol);
++nPasses2;
}
// std::cerr<<" "<<field2->calcEnergy()<<" after npasses2:
// "<<nPasses2<<std::endl;
}
return true;
}
// the minimization using experimental torsion angle preferences
bool minimizeWithExpTorsions(RDGeom::PointPtrVect &positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams) {
PRECONDITION(eargs.etkdgDetails, "bogus etkdgDetails pointer");
bool planar = true;
// convert to 3D positions and create coordMap
RDGeom::Point3DPtrVect positions3D;
for (auto &position : positions) {
positions3D.push_back(
new RDGeom::Point3D((*position)[0], (*position)[1], (*position)[2]));
}
// create the force field
std::unique_ptr<ForceFields::ForceField> field;
if (embedParams.useBasicKnowledge) { // ETKDG or KDG
field.reset(DistGeom::construct3DForceField(
*eargs.mmat, positions3D, *eargs.etkdgDetails));
} else { // plain ETDG
field.reset(DistGeom::constructPlain3DForceField(
*eargs.mmat, positions3D, *eargs.etkdgDetails));
}
// minimize!
field->initialize();
if (field->calcEnergy() > ERROR_TOL) {
// while (needMore) {
field->minimize(300, embedParams.optimizerForceTol);
// ++nPasses;
//}
}
// std::cout << field->calcEnergy() << std::endl;
// check for planarity if ETKDG or KDG
if (embedParams.useBasicKnowledge) {
// create a force field with only the impropers
std::unique_ptr<ForceFields::ForceField> field2(
DistGeom::construct3DImproperForceField(
*eargs.mmat, positions3D, eargs.etkdgDetails->improperAtoms,
eargs.etkdgDetails->atomNums));
field2->initialize();
// check if the energy is low enough
double planarityTolerance = 0.7;
if (field2->calcEnergy() >
eargs.etkdgDetails->improperAtoms.size() * planarityTolerance) {
#ifdef DEBUG_EMBEDDING
std::cerr << " planar fail: " << field2->calcEnergy() << " "
<< eargs.etkdgDetails->improperAtoms.size() * planarityTolerance
<< std::endl;
#endif
planar = false;
}
}
// overwrite positions and delete the 3D ones
for (unsigned int i = 0; i < positions3D.size(); ++i) {
(*positions[i])[0] = (*positions3D[i])[0];
(*positions[i])[1] = (*positions3D[i])[1];
(*positions[i])[2] = (*positions3D[i])[2];
delete positions3D[i];
}
return planar;
}
bool finalChiralChecks(RDGeom::PointPtrVect *positions,
const detail::EmbedArgs &eargs,
const EmbedParameters &embedParams) {
RDUNUSED_PARAM(embedParams);
// "distance matrix" chirality test
std::set<int> atoms;
for (const auto chiralSet : *eargs.chiralCenters) {
if (chiralSet->d_idx0 != chiralSet->d_idx4) {
atoms.insert(chiralSet->d_idx0);
atoms.insert(chiralSet->d_idx1);
atoms.insert(chiralSet->d_idx2);
atoms.insert(chiralSet->d_idx3);
atoms.insert(chiralSet->d_idx4);
}
}
std::vector<int> atomsToCheck(atoms.begin(), atoms.end());
if (atomsToCheck.size() > 0) {
if (!_boundsFulfilled(atomsToCheck, *eargs.mmat, *positions)) {
#ifdef DEBUG_EMBEDDING
std::cerr << " fail3a! (" << atomsToCheck[0] << ") iter: " << iter
<< std::endl;
#endif
return false;
}
}
// "center in volume" chirality test
for (const auto chiralSet : *eargs.chiralCenters) {
// it could happen that the centroid is outside the volume defined
// by the other four points. That is also a fail.
if (!_centerInVolume(chiralSet, *positions)) {
#ifdef DEBUG_EMBEDDING
std::cerr << " fail3b! (" << chiralSet->d_idx0 << ") iter: " << iter
<< std::endl;
#endif
return false;
}
}
return true;
}
bool embedPoints(RDGeom::PointPtrVect *positions, detail::EmbedArgs eargs,
EmbedParameters embedParams, int seed) {
PRECONDITION(positions, "bogus positions");
if (embedParams.maxIterations == 0) {
embedParams.maxIterations = 10 * positions->size();
}
RDNumeric::DoubleSymmMatrix distMat(positions->size(), 0.0);
// The basin threshold just gets us into trouble when we're using
// random coordinates since it ends up ignoring 1-4 (and higher)
// interactions. This causes us to get folded-up (and self-penetrating)
// conformations for large flexible molecules
if (embedParams.useRandomCoords) embedParams.basinThresh = 1e8;
RDKit::double_source_type *rng = nullptr;
RDKit::rng_type *generator;
RDKit::uniform_double *distrib;
CHECK_INVARIANT(seed >= -1,
"random seed must either be positive, zero, or negative one");
if (seed > -1) {
generator = new RDKit::rng_type(42u);
generator->seed(seed);
distrib = new RDKit::uniform_double(0.0, 1.0);
rng = new RDKit::double_source_type(*generator, *distrib);
} else {
rng = &RDKit::getDoubleRandomSource();
}
bool gotCoords = false;
unsigned int iter = 0;
while ((gotCoords == false) && (iter < embedParams.maxIterations)) {
++iter;
gotCoords = EmbeddingOps::generateInitialCoords(positions, eargs,
embedParams, distMat, rng);
#ifdef DEBUG_EMBEDDING
if (!gotCoords) {
std::cerr << "Initial embedding failed!, Iter: " << iter << std::endl;
}
#endif
if (gotCoords) {
gotCoords =
EmbeddingOps::firstMinimization(positions, eargs, embedParams);
if (gotCoords) {
gotCoords = EmbeddingOps::checkTetrahedralCenters(positions, eargs,
embedParams);
}
// Check if any of our chiral centers are badly out of whack.
if (gotCoords && embedParams.enforceChirality &&
eargs.chiralCenters->size() > 0) {
gotCoords =
EmbeddingOps::checkChiralCenters(positions, eargs, embedParams);
}
// redo the minimization if we have a chiral center
// or have started from random coords.
if (gotCoords &&
(eargs.chiralCenters->size() > 0 || embedParams.useRandomCoords)) {
gotCoords = EmbeddingOps::minimizeFourthDimension(positions, eargs,
embedParams);
}
// (ET)(K)DG
if (gotCoords && (embedParams.useExpTorsionAnglePrefs ||
embedParams.useBasicKnowledge)) {
gotCoords = EmbeddingOps::minimizeWithExpTorsions(*positions, eargs,
embedParams);
}
// test if chirality is correct
if (embedParams.enforceChirality && gotCoords &&
(eargs.chiralCenters->size() > 0)) {
gotCoords =
EmbeddingOps::finalChiralChecks(positions, eargs, embedParams);
}
} // if(gotCoords)
} // while
if (seed > -1 && rng) {
delete rng;
delete generator;
delete distrib;
}
return gotCoords;
}
void findChiralSets(const ROMol &mol, DistGeom::VECT_CHIRALSET &chiralCenters,
DistGeom::VECT_CHIRALSET &tetrahedralCenters,
const std::map<int, RDGeom::Point3D> *coordMap) {
for (const auto &atom : mol.atoms()) {
if (atom->getAtomicNum() != 1) { // skip hydrogens
Atom::ChiralType chiralType = atom->getChiralTag();
if ((chiralType == Atom::CHI_TETRAHEDRAL_CW ||
chiralType == Atom::CHI_TETRAHEDRAL_CCW) ||
((atom->getAtomicNum() == 6 || atom->getAtomicNum() == 7) &&
atom->getDegree() == 4)) {
// make a chiral set from the neighbors
INT_VECT nbrs;
nbrs.reserve(4);
// find the neighbors of this atom and enter them into the
// nbr list
ROMol::OEDGE_ITER beg, end;
boost::tie(beg, end) = mol.getAtomBonds(atom);
while (beg != end) {
nbrs.push_back(mol[*beg]->getOtherAtom(atom)->getIdx());
++beg;
}
// if we have less than 4 heavy atoms as neighbors,
// we need to include the chiral center into the mix
// we should at least have 3 though
bool includeSelf = false;
RDUNUSED_PARAM(includeSelf);
CHECK_INVARIANT(nbrs.size() >= 3, "Cannot be a chiral center");
if (nbrs.size() < 4) {
nbrs.insert(nbrs.end(), atom->getIdx());
includeSelf = true;
}
// now create a chiral set and set the upper and lower bound on the
// volume
if (chiralType == Atom::CHI_TETRAHEDRAL_CCW) {
// postive chiral volume
auto *cset = new DistGeom::ChiralSet(atom->getIdx(), nbrs[0], nbrs[1],
nbrs[2], nbrs[3], 5.0, 100.0);
DistGeom::ChiralSetPtr cptr(cset);
chiralCenters.push_back(cptr);
} else if (chiralType == Atom::CHI_TETRAHEDRAL_CW) {
auto *cset = new DistGeom::ChiralSet(atom->getIdx(), nbrs[0], nbrs[1],
nbrs[2], nbrs[3], -100.0, -5.0);
DistGeom::ChiralSetPtr cptr(cset);
chiralCenters.push_back(cptr);
} else {
if ((coordMap && coordMap->find(atom->getIdx()) != coordMap->end()) ||
(mol.getRingInfo()->isInitialized() &&
(mol.getRingInfo()->numAtomRings(atom->getIdx()) < 2 ||
mol.getRingInfo()->isAtomInRingOfSize(atom->getIdx(), 3)))) {
// we only want to these tests for ring atoms that are not part of
// the coordMap
// there's no sense doing 3-rings because those are a nightmare
} else {
auto *cset = new DistGeom::ChiralSet(
atom->getIdx(), nbrs[0], nbrs[1], nbrs[2], nbrs[3], 0.0, 0.0);
DistGeom::ChiralSetPtr cptr(cset);
tetrahedralCenters.push_back(cptr);
}
}
} // if block -chirality check
} // if block - heavy atom check
} // for loop over atoms
} // end of _findChiralSets
void adjustBoundsMatFromCoordMap(
DistGeom::BoundsMatPtr mmat, unsigned int nAtoms,
const std::map<int, RDGeom::Point3D> *coordMap) {
RDUNUSED_PARAM(nAtoms);
for (auto iIt = coordMap->begin(); iIt != coordMap->end(); ++iIt) {
unsigned int iIdx = iIt->first;
const RDGeom::Point3D &iPoint = iIt->second;
auto jIt = iIt;
while (++jIt != coordMap->end()) {
unsigned int jIdx = jIt->first;
const RDGeom::Point3D &jPoint = jIt->second;
double dist = (iPoint - jPoint).length();
mmat->setUpperBound(iIdx, jIdx, dist);
mmat->setLowerBound(iIdx, jIdx, dist);
}
}
}
bool setupInitialBoundsMatrix(
ROMol *mol, DistGeom::BoundsMatPtr mmat,
const std::map<int, RDGeom::Point3D> *coordMap,
const EmbedParameters &params,
ForceFields::CrystalFF::CrystalFFDetails &etkdgDetails) {
PRECONDITION(mol, "bad molecule");
unsigned int nAtoms = mol->getNumAtoms();
if (params.useExpTorsionAnglePrefs || params.useBasicKnowledge) {
ForceFields::CrystalFF::getExperimentalTorsions(
*mol, etkdgDetails, params.useExpTorsionAnglePrefs,
params.useBasicKnowledge, params.ETversion, params.verbose);
setTopolBounds(*mol, mmat, etkdgDetails.bonds, etkdgDetails.angles, true,
false);
etkdgDetails.atomNums.resize(nAtoms);
for (unsigned int i = 0; i < nAtoms; ++i) {
etkdgDetails.atomNums[i] = (*mol).getAtomWithIdx(i)->getAtomicNum();
}
} else {
setTopolBounds(*mol, mmat, true, false);
}
double tol = 0.0;
if (coordMap) {
adjustBoundsMatFromCoordMap(mmat, nAtoms, coordMap);
tol = 0.05;
}
if (!DistGeom::triangleSmoothBounds(mmat, tol)) {
// ok this bound matrix failed to triangle smooth - re-compute the
// bounds matrix without 15 bounds and with VDW scaling
initBoundsMat(mmat);
setTopolBounds(*mol, mmat, false, true);
if (coordMap) {
adjustBoundsMatFromCoordMap(mmat, nAtoms, coordMap);
}
// try triangle smoothing again
if (!DistGeom::triangleSmoothBounds(mmat, tol)) {
// ok, we're not going to be able to smooth this,
if (params.ignoreSmoothingFailures) {
// proceed anyway with the more relaxed bounds matrix
initBoundsMat(mmat);
setTopolBounds(*mol, mmat, false, true);
if (coordMap) {
adjustBoundsMatFromCoordMap(mmat, nAtoms, coordMap);
}
} else {
BOOST_LOG(rdWarningLog)
<< "Could not triangle bounds smooth molecule." << std::endl;
return false;
}
}
}
return true;
}
} // namespace EmbeddingOps
void _fillAtomPositions(RDGeom::Point3DConstPtrVect &pts, const Conformer &conf,
const ROMol &mol, bool onlyHeavyAtomsForRMS) {
unsigned int na = conf.getNumAtoms();
pts.clear();
pts.reserve(na);
for (const auto &atom : mol.atoms()) {
// FIX: should we include D and T here?
if (onlyHeavyAtomsForRMS && atom->getAtomicNum() == 1) {
continue;
}
pts.push_back(&conf.getAtomPos(atom->getIdx()));
}
}
bool _isConfFarFromRest(const ROMol &mol, const Conformer &conf,
double threshold, bool onlyHeavyAtomsForRMS) {
// NOTE: it is tempting to use some triangle inequality to prune
// conformations here but some basic testing has shown very
// little advantage and given that the time for pruning fades in
// comparison to embedding - we will use a simple for loop below
// over all conformation until we find a match
RDGeom::Point3DConstPtrVect refPoints, prbPoints;
_fillAtomPositions(refPoints, conf, mol, onlyHeavyAtomsForRMS);
double ssrThres = conf.getNumAtoms() * threshold * threshold;
for (ROMol::ConstConformerIterator confi = mol.beginConformers();
confi != mol.endConformers(); ++confi) {
_fillAtomPositions(prbPoints, *(*confi), mol, onlyHeavyAtomsForRMS);
RDGeom::Transform3D trans;
auto ssr = RDNumeric::Alignments::AlignPoints(refPoints, prbPoints, trans);
if (ssr < ssrThres) {
return false;
}
}
return true;
}
namespace detail {
template <class T>
bool multiplication_overflows_(T a, T b) {
// a * b > c if and only if a > c / b
if (a == 0 || b == 0) return false;
if (a > std::numeric_limits<T>::max() / b) return true;
return false;
}
void embedHelper_(int threadId, int numThreads, EmbedArgs *eargs,
const EmbedParameters *params) {
PRECONDITION(eargs, "bogus eargs");
PRECONDITION(params, "bogus params");
unsigned int nAtoms = eargs->mmat->numRows();
RDGeom::PointPtrVect positions(nAtoms);
for (unsigned int i = 0; i < nAtoms; ++i) {
if (eargs->fourD) {
positions[i] = new RDGeom::PointND(4);
} else {
positions[i] = new RDGeom::Point3D();
}
}
for (size_t ci = 0; ci < eargs->confs->size(); ci++) {
if (rdcast<int>(ci % numThreads) != threadId) continue;
if (!(*eargs->confsOk)[ci]) {
// we call this function for each fragment in a molecule,
// if one of the fragments has already failed, there's no
// sense in embedding this one
continue;
}
CHECK_INVARIANT(
params->randomSeed >= -1,
"random seed must either be positive, zero, or negative one");
int new_seed = params->randomSeed;
if (new_seed > -1) {
if (!multiplication_overflows_(rdcast<int>(ci + 1), params->randomSeed)) {
// old method of computing a new seed
new_seed = (ci + 1) * params->randomSeed;
} else {
// If the above simple multiplication will overflow, use a
// cheap and easy way to hash the conformer index and seed
// together: for N'ary numerical system, where N is the
// maximum possible value of the pair of numbers. The
// following will generate unique integers:
// hash(a, b) = a + b * N
size_t big_seed = rdcast<size_t>(params->randomSeed);
size_t max_val = std::max(ci + 1, big_seed);
size_t big_num = big_seed + max_val * (ci + 1);
// only grab the first 31 bits xor'd with the next 31 bits to
// make sure its positive, careful, the 'ULL' is important
// here, 0x7fffffff is the 'int' type because of C default
// number semantics and that we definitely don't want!
const size_t positive_int_mask = 0x7fffffffULL;
size_t folded_num = (big_num & positive_int_mask) ^ (big_num >> 31ULL);
new_seed = rdcast<int>(folded_num & positive_int_mask);
}
}
CHECK_INVARIANT(new_seed >= -1,
"Something went wrong calculating a new seed");
bool gotCoords =
EmbeddingOps::embedPoints(&positions, *eargs, *params, new_seed);
// copy the coordinates into the correct conformer
if (gotCoords) {
Conformer *conf = (*eargs->confs)[ci];
unsigned int fragAtomIdx = 0;
for (unsigned int i = 0; i < conf->getNumAtoms(); ++i) {
if (!eargs->fragMapping ||
(*eargs->fragMapping)[i] == static_cast<int>(eargs->fragIdx)) {
conf->setAtomPos(i, RDGeom::Point3D((*positions[fragAtomIdx])[0],
(*positions[fragAtomIdx])[1],
(*positions[fragAtomIdx])[2]));
++fragAtomIdx;
}
}
} else {
(*eargs->confsOk)[ci] = 0;
}
}
for (unsigned int i = 0; i < nAtoms; ++i) {
delete positions[i];
}
}
} // end of namespace detail
void EmbedMultipleConfs(ROMol &mol, INT_VECT &res, unsigned int numConfs,
const EmbedParameters &params) {
if (!mol.getNumAtoms()) {
throw ValueErrorException("molecule has no atoms");
}
if (params.ETversion < 1 || params.ETversion > 2) {
throw ValueErrorException(
"Only version 1 and 2 of the experimental "
"torsion-angle preferences (ETversion) supported");
}
// initialize the conformers we're going to be creating:
if (params.clearConfs) {
res.clear();
mol.clearConformers();
}
std::vector<Conformer *> confs(numConfs);
for (unsigned int i = 0; i < numConfs; ++i) {
confs[i] = new Conformer(mol.getNumAtoms());
}
boost::dynamic_bitset<> confsOk(numConfs);
confsOk.set();
INT_VECT fragMapping;
std::vector<ROMOL_SPTR> molFrags =
MolOps::getMolFrags(mol, true, &fragMapping);
const std::map<int, RDGeom::Point3D> *coordMap = params.coordMap;
if (molFrags.size() > 1 && coordMap) {
BOOST_LOG(rdWarningLog)
<< "Constrained conformer generation (via the coordMap argument) "
"does not work with molecules that have multiple fragments."
<< std::endl;
coordMap = nullptr;
}
// we will generate conformations for each fragment in the molecule
// separately, so loop over them:
for (unsigned int fragIdx = 0; fragIdx < molFrags.size(); ++fragIdx) {
ROMOL_SPTR piece = molFrags[fragIdx];
unsigned int nAtoms = piece->getNumAtoms();
// create and initialize the distance bounds matrix
DistGeom::BoundsMatPtr mmat(new DistGeom::BoundsMatrix(nAtoms));
initBoundsMat(mmat);
ForceFields::CrystalFF::CrystalFFDetails etkdgDetails;
if (!EmbeddingOps::setupInitialBoundsMatrix(piece.get(), mmat, coordMap,
params, etkdgDetails)) {
// return if we couldn't setup the bounds matrix
// possible causes include a triangle smoothing failure
return;
}
// find all the chiral centers in the molecule
MolOps::assignStereochemistry(*piece);
DistGeom::VECT_CHIRALSET chiralCenters;
DistGeom::VECT_CHIRALSET tetrahedralCarbons;
EmbeddingOps::findChiralSets(*piece, chiralCenters, tetrahedralCarbons,
coordMap);
// if we have any chiral centers or are using random coordinates, we will
// first embed the molecule in four dimensions, otherwise we will use 3D
bool fourD = false;
if (params.useRandomCoords || chiralCenters.size() > 0) {
fourD = true;
}
int numThreads = getNumThreadsToUse(params.numThreads);
// do the embedding, using multiple threads if requested
detail::EmbedArgs eargs = {
&confsOk, fourD, &fragMapping, &confs, fragIdx,
mmat, &chiralCenters, &tetrahedralCarbons, &etkdgDetails};
if (numThreads == 1) {
detail::embedHelper_(0, 1, &eargs, &params);
}
#ifdef RDK_THREADSAFE_SSS
else {
std::vector<std::future<void>> tg;
for (int tid = 0; tid < numThreads; ++tid) {
tg.emplace_back(std::async(std::launch::async, detail::embedHelper_,
tid, numThreads, &eargs, &params));
}
for (auto &fut : tg) {
fut.get();
}
}
#endif
}
for (unsigned int ci = 0; ci < confs.size(); ++ci) {
Conformer *conf = confs[ci];
if (confsOk[ci]) {
// check if we are pruning away conformations and
// a closeby conformation has already been chosen :
if (params.pruneRmsThresh > 0.0 &&
!_isConfFarFromRest(mol, *conf, params.pruneRmsThresh,
params.onlyHeavyAtomsForRMS)) {
delete conf;
} else {
int confId = (int)mol.addConformer(conf, true);
res.push_back(confId);
}
} else {
delete conf;
}
}
}
} // end of namespace DGeomHelpers
} // end of namespace RDKit
Reference in New Issue View Git Blame Copy Permalink