In my project, I study reaction computationally. The initial step in my workflow involves a conformational search using a constraint to bring the two reactive parts on my molecule together (reaction forms ring).

This conformational searching generates many confirmations, and it can be difficult to tell which are different.

Previously I use redundant comformer search which narrows down before I pick which conformers to study by DFT but today it was suggested to use a clustering method to group the conformers. In the suggestion i was told that the clustering would group into chair vs boat vs twist etc which would be useful. I tried this using maestros inbuilt algorithm but the results were not very good - all of the clusters contained multiple ring types.

Is there any better way of clustering in order to sort my results. Using external software for instance, or mathematically. I also wonder whether both are needed (redundant conformer elimination then clustering, or just cluster all results).

  • 1
    $\begingroup$ Welcome to the site! Great first question. If you haven't already, take the tour and if you have any other questions about the site check at the help center. $\endgroup$ – Tyberius Feb 23 '18 at 18:21
  • $\begingroup$ I'll preface this by saying I'm sure there are much better ways to do this, but a simple approach could be to DFT-optimize the structure of each group (1 boat, 1 chair, etc.) and then calculate the root-mean-square deviation in atomic positions of all the generated conformers from each of these prototypical structures. I'm sure Open Babel has some better method though, so this is just food for thought. (Tangentially related, but this seems like a good problem for machine learning-based multiclass classification.) $\endgroup$ – Argon Feb 24 '18 at 1:53
  • $\begingroup$ Thanks for commenting, but this doesn’t really simplify things at all - my current approach manually going through result gives this. I’m interested in clustering to do the sorting rather that sorting to be able to clusterering. $\endgroup$ – Ben Feb 24 '18 at 10:07
  • $\begingroup$ What type of clustering algorithm are you using? Most clustering algorithms have hyper-parameters, which influence which clusters are found and how many. $\endgroup$ – Deathbreath Mar 2 '18 at 15:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.