Code/ForceField/MMFF/Params.h, Code/ForceField/UFF/Params.h,
Code/GraphMol/ForceFieldHelpers/MMFF/AtomTyper.cpp
and Code/GraphMol/ForceFieldHelpers/MMFF/AtomTyper.h (I realized
their uselessness thanks to a warning issued by Intel C++ compiler)
- refactored O3A code
- added the possibility to set weighted constraints on selected
atom pairs
- added an option to carry out local-only optimization
data races in the multithreaded test
- Removed a spurious #include in Code/ForceField/Wrap/ForceField.cpp
- Restored caching in Code/GraphMol/Descriptors/Crippen.cpp
and scoring functions
- Added cost, weight and scoring functions using atom-based
Crippen logP contributions
- Added relevant tests for the new functionality
Important points:
-----------------
- The constructors now accept optional pointers to
MolHistogram objects instead of pointers to arrays of
double. I think it is better for performance, since
rebuilding the histogram involves running through two
nested loops over all atoms, even though the
3DDistanceMat is provided to the O3A constructor.
This change breaks binary compatibility with previous
C++ programs linking to RDKit MolAlign library; I do not
think this is big issue, and while I was there I made
other changes which cause binary incompatibility.
If needed, we may preserve binary compatibility
reverting the MolHistogram change and the other ones.
- From Python, nothing changed in the interface to previous
MMFFO3A functionality. It MIGHT be more appropriate to
change the "GetO3A()" function into "GetMMFFO3A()", but
I have not done that to avoid breaking existing scripts;
the choice is yours
- As of now, the code contains a number of conditional
compilation directives checking for the
USE_O3A_CONSTRUCTOR macro; if USE_O3A_CONSTRUCTOR is
defined, then the code is built with an alternative O3A
constructor which allows choosing whether one wishes
to use MMFF or Crippen descriptors. Otherwise, no
alternative constructor is built, but rather two
functions which return a pointer to an O3A object.
I prefer by far the first solution (namely, with
USE_O3A_CONSTRUCTOR defined), but again, the choice is
yours. The code is tested and works in both cases.
- Custom cost, weight and scoring functions can easily
be defined in external programs without need to rebuild
the RDKit, allowing flexibility. The new custom
functionality can be accessed by calling the "bigger"
O3A constructor.
computed in the usual way, while previously the value computed
by MolAlign::AlignMol(), which incorporates weights, was returned
- the C++ and Python test results were updated accordingly
