The chiral validity MCS filter added in #1356 is prohibitively slow in some important edge cases where the MCS search visits a large number of chirally-invalid leaf nodes before finding a valid one, because each invocation of the leaf filter is expensive.
One opportunity for improvement is to remove some of the substantial redundant work that is done when setting up chiral restraints for the hybrid mol in the leaf filter. Much of the expensive work is independent of the dummy group assignment, and can be precomputed outside of the MCS search.
This PR refactors bond and chiral restraint setup to reduce the time for a single execution of the chiral validity filter from 200-300 ms down to ~10 ms.
Summary of changes
Precompute quantities independent of core and dummy group assignment during MCS initialization
Move assertion that hybrid mol is connected to setup_end_state
Refactor setup_dummy_bond_and_chiral_interactions using vectorized numpy ops
Remove recursive_map, replace with array indexing
Vectorize bond and chiral atom ixn canonicalization
Move sanity-check assertions behind verify flag (defaulting to True; disabled during MCS)
The chiral validity MCS filter added in #1356 is prohibitively slow in some important edge cases where the MCS search visits a large number of chirally-invalid leaf nodes before finding a valid one, because each invocation of the leaf filter is expensive.
One opportunity for improvement is to remove some of the substantial redundant work that is done when setting up chiral restraints for the hybrid mol in the leaf filter. Much of the expensive work is independent of the dummy group assignment, and can be precomputed outside of the MCS search.
This PR refactors bond and chiral restraint setup to reduce the time for a single execution of the chiral validity filter from 200-300 ms down to ~10 ms.
Summary of changes
setup_end_statesetup_dummy_bond_and_chiral_interactionsusing vectorized numpy opsrecursive_map, replace with array indexingverifyflag (defaulting to True; disabled during MCS)