Files
rdkit/Code/GraphMol/Subgraphs/SubgraphUtils.cpp
Brian Kelley 70540c2eed Add extract mol fragment api (#8811)
* Create a function to extract some specified atoms from a ROMol as a new ROMol by creating new graph (#8742)

This adds a new api, `RDKit::MolOps::ExtractMolFragment`, to allow efficient
extractions of mol fragments from large mols. Compared to the approach where
we delete "unwanted" atoms/bonds from the input mol, this api is faster for
small mols (about 2x faster) and at least 3x faster for big mols
(was 10x faster for "CCC"*1000).

* clang-format

* review comments

* cleanup

* Consolidate copying subsets of molecules

* Readd missing tests

* Update comment to restart build

* Remove missing test

* Remove debugging comment, fix warnings

* Fix warnings on gcc11

* Add docs

* Make vector<bool> dynamic_bitset<>

* Update copyright

* Add swig wrappers

* Use new designated constructor API

* Fix windows builds

* Change enum values from unsigned int to integer

* Fix unsigned int variable

* Update Code/GraphMol/Wrap/test_subset.py

Co-authored-by: Greg Landrum <greg.landrum@gmail.com>

* Update Code/GraphMol/Subset.cpp

Co-authored-by: Greg Landrum <greg.landrum@gmail.com>

* Update Code/JavaWrappers/gmwrapper/src-test/org/RDKit/ChemTransformsTests.java

Co-authored-by: Greg Landrum <greg.landrum@gmail.com>

* Reponse to review

* Fix documentation

* Remove comments

* Remove unnecessary comments

* Fix one liners

* Change assertion to be clearer (and not one-liners)

* Run clang-format

---------

Co-authored-by: Your Name <you@example.com>
Co-authored-by: Hussein Faara <hussein.faara@schrodinger.com>
Co-authored-by: Brian Kelley <bkelley@glysade.com>
Co-authored-by: Greg Landrum <greg.landrum@gmail.com>
2025-12-09 15:06:29 +01:00

180 lines
6.1 KiB
C++

// $Id$
//
// Copyright (C) 2003-2013 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.
//
#include "SubgraphUtils.h"
#include "Subgraphs.h"
#include <RDGeneral/utils.h>
#include <GraphMol/RDKitBase.h>
#include <GraphMol/Subset.h>
#include <GraphMol/QueryAtom.h>
#include <GraphMol/QueryBond.h>
#include <algorithm>
#include <iostream>
#include <map>
#include <RDGeneral/hash/hash.hpp>
namespace RDKit {
namespace Subgraphs {
ROMol *pathToSubmol(const ROMol &mol, const PATH_TYPE &path, bool useQuery) {
INT_MAP_INT aIdxMap;
return pathToSubmol(mol, path, useQuery, aIdxMap);
}
ROMol *pathToSubmol(const ROMol &mol, const PATH_TYPE &path, bool useQuery,
INT_MAP_INT &atomIdxMap) {
SubsetOptions options;
options.copyAsQuery = useQuery;
options.copyCoordinates = true;
options.method = SubsetMethod::BONDS;
std::vector<unsigned int> upath{path.begin(), path.end()};
SubsetInfo subsetInfo;
auto res = copyMolSubset(mol, upath, subsetInfo, options);
atomIdxMap.clear();
for (auto mapping : subsetInfo.atomMapping) {
atomIdxMap[mapping.first] = mapping.second;
}
return res.release();
}
PATH_TYPE bondListFromAtomList(const ROMol &mol, const PATH_TYPE &atomIds) {
PATH_TYPE bids;
unsigned int natms = atomIds.size();
if (natms <= 1) {
return bids; // FIX: should probably throw an exception
}
for (unsigned int i = 0; i < natms; i++) {
for (unsigned int j = i + 1; j < natms; j++) {
const Bond *bnd = mol.getBondBetweenAtoms(atomIds[i], atomIds[j]);
if (bnd) {
int bid = bnd->getIdx();
bids.push_back(bid);
}
}
}
return bids;
}
using std::int32_t;
using std::uint32_t;
DiscrimTuple calcPathDiscriminators(const ROMol &mol, const PATH_TYPE &path,
bool useBO,
std::vector<std::uint32_t> *extraInvars) {
if (extraInvars) {
CHECK_INVARIANT(extraInvars->size() == mol.getNumAtoms(),
"bad extra invars");
}
// Start by collecting the atoms in the path and their degrees
std::vector<int32_t> atomsUsed(mol.getNumAtoms(),
-1); // map from atom index->path index
std::vector<const Atom *> atoms; // to contain the atoms in the path
std::vector<uint32_t> pathDegrees; // degrees of each atom *in the path*
for (int pathIter : path) {
const Bond *bond = mol.getBondWithIdx(pathIter);
if (atomsUsed[bond->getBeginAtomIdx()] < 0) {
atomsUsed[bond->getBeginAtomIdx()] = static_cast<int>(atoms.size());
atoms.push_back(bond->getBeginAtom());
pathDegrees.push_back(1);
} else {
pathDegrees[atomsUsed[bond->getBeginAtomIdx()]] += 1;
}
if (atomsUsed[bond->getEndAtomIdx()] < 0) {
atomsUsed[bond->getEndAtomIdx()] = static_cast<int>(atoms.size());
atoms.push_back(bond->getEndAtom());
pathDegrees.push_back(1);
} else {
pathDegrees[atomsUsed[bond->getEndAtomIdx()]] += 1;
}
}
// Calculate the atomic invariants
unsigned int nAtoms = atoms.size();
std::vector<uint32_t> invars(nAtoms);
for (unsigned int i = 0; i < nAtoms; ++i) {
const Atom *atom = atoms[i];
uint32_t invar = atom->getAtomicNum();
gboost::hash_combine(invar, pathDegrees[i]);
gboost::hash_combine(invar, atom->getFormalCharge());
int deltaMass = static_cast<int>(
atom->getMass() -
PeriodicTable::getTable()->getAtomicWeight(atom->getAtomicNum()));
gboost::hash_combine(invar, deltaMass);
if (atom->getIsAromatic()) {
gboost::hash_combine(invar, 1);
}
if (extraInvars) {
gboost::hash_combine(invar, (*extraInvars)[atom->getIdx()]);
}
invars[i] = invar;
}
// now do the Morgan iterations:
// the most number of cycles we need for the atoms on the edges
// to feel each other is pathSize/2
// EFF: it may be worth revisiting this at some point to see
// if the iteration count can be even smaller (and if it
// makes a difference in runtime)
unsigned int nCycles = path.size() / 2 + 1;
gboost::hash<std::vector<uint32_t>> vectHasher;
for (unsigned int cycle = 0; cycle < nCycles; ++cycle) {
// let each atom feel it's neighbors:
std::vector<std::vector<uint32_t>> locInvars(nAtoms);
for (int pathIter : path) {
const Bond *bond = mol.getBondWithIdx(pathIter);
uint32_t v1 = invars[atomsUsed[bond->getBeginAtomIdx()]];
uint32_t v2 = invars[atomsUsed[bond->getEndAtomIdx()]];
if (useBO) {
gboost::hash_combine(v1, static_cast<uint32_t>(bond->getBondType()));
gboost::hash_combine(v2, static_cast<uint32_t>(bond->getBondType()));
}
locInvars[atomsUsed[bond->getBeginAtomIdx()]].push_back(v2);
locInvars[atomsUsed[bond->getEndAtomIdx()]].push_back(v1);
}
// we need to sort by the neighbor invariants to be order
// independent:
for (unsigned int i = 0; i < nAtoms; ++i) {
std::sort(locInvars[i].begin(), locInvars[i].end());
invars[i] = vectHasher(locInvars[i]);
}
}
// again, a sort for order independence:
std::sort(invars.begin(), invars.end());
uint32_t pathInvar = vectHasher(invars);
// also include the path size (bond count) and number of atoms
// in the discriminator
return std::make_tuple(pathInvar, path.size(), nAtoms);
}
//
// This is intended for use on either subgraphs or paths.
// The entries in PATH_LIST should refer to bonds though (not
// atoms)
//
PATH_LIST uniquifyPaths(const ROMol &mol, const PATH_LIST &allPaths,
bool useBO) {
PATH_LIST res;
std::vector<DiscrimTuple> discrimsSeen;
for (const auto &allPath : allPaths) {
DiscrimTuple discrims = calcPathDiscriminators(mol, allPath, useBO);
if (std::find(discrimsSeen.begin(), discrimsSeen.end(), discrims) ==
discrimsSeen.end()) {
discrimsSeen.push_back(discrims);
res.push_back(allPath);
}
}
return res;
}
} // end of namespace Subgraphs
} // end of namespace RDKit