Conclusions and outlook

We presented an overview on structural superpositioning and molecular similarity, in general. We contributed to this field the alignment tool FlexS which allows to flexibly superimpose one structure onto a rigid reference rapidly. Several recent enhancements of the software, like full automation, fragment-based superpositioning at high speed, and enriching the representation of the reference structure, made our software applicable to screening virtual databases.

We designed a two-step filtering protocol that enables us to go through large compound collections and mimic real-world applications in our evaluation scenario. It is demonstrated that the critical parameters, like the selection of an appropriate reference compound, can be trained on a small-size training set at low computational costs and that results carry over reasonably well to the large-scale application.

A topic to be addressed in future work is multiple flexible superposition considering several compounds simultaneously. Iterated pairwise alignments and making use of the merging of Gaussian representations of several molecules certainly can produce reasonable starting positions for a multi-molecule flexible post-optimization on which we are currently working.

Another important issue is that with the currently used screening protocol we reduce the wealth of information contained in the laboriously aligned ligands to a single number, namely the similarity score. There is certainly much more information present in an alignment that could be exploited to predict activity.

0 0

Post a comment