Atom Mapping Scoring

Atom mapping scorers try to evaluate the quality of a given mapping. This however is a non-trivial task as the quality of a mapping, depends on the use case of the mapping. A simple case example could be, if a program, for any reason, reindexes a molecule without changing the conformation, an atom mapping can be used to get an index matching of before and after the program usage. Usually, in such a case, you want to map as many atoms as possible with a distance of 0 angstroms to each other.

A less trivial example is the atom mapping case for molecule transformations like in hybrid topology free energy calculation approaches. In such a case the mapping quality depends on the likelihood of the transformation to converge well and to give a reasonable free energy. This in a way would require the true and calculated result of the calculation and lead to a henn egg problem. To tackle this problem usually simplistic approaches are used to estimate the success likelihood of the transformation. Criteria for rule-based approaches could be element changes, mapped atom displacement, flexibility changes or polarity changes, appearing atom number; basically, anything that introduces a difference from one molecule to the other leading to an increase of the perturbation. Keep in mind, that this does not only depend on the ligands themselves but also the environment interacting with those ligands.

Additionally one can start adding parameter-specific scores, that depend on force fields or other method-related aspects. However this might theoretically improve the method outcome, it could hinder the transferability of the scorer from one method to another (overfitting).

In Katograf we added some functionality, that can be used as an aspect of an atom mapping scorer, like the MappingRMSDScorer, checking the displacement of atoms by the scorers or the MappingVolumeRatioScorer, checking the volume overlap of the two molecules.