3 heavy atoms or less

[kvkarandashev]

Is there a fundamental reason to why molecules with less than 4 atoms get a special treatment? I think it was mentioned that this is because of dgsol, but if it's the problem can't we write a code of our own which would solve the geometry problem for 1, 2, and 3 atoms?

[Dom1L]

I'm actually not 100% sure anymore whether it was due to the heavy atom embedding or if it was some complication with my hydrogen code.

I would have to look into this again.

Do you need to work with such small molecules?

I dont see any problems with custom code for that, the 1 or 2 atoms case are easy, the 3 atom one is the one that needs some hacking. Since I already switched the distance geometry solver in the dev branch for another one (lsbuild) that is written in Python it could be doable. I dont remember exactly whether dgsol actually crashed for 3 distances or not.

[kvkarandashev]

If you look at the create_model.py example that I've sent related to issue #2, at line 60 you'll see I needed to kick molecules from 3 atoms or less, otherwise there will be a conflict due to some molecule that does not undergo heavy atom filtering generating a representations array of length 9 and trying to fit into into a padded representations array of length 7. I suspect that the best way to solve the issue is to get rid of that filtering exception altogether.

[ferchault]

The hydrogen saturation likely is the issue, since they are triangulated. However for small molecules, this can be answered from pure geometry alone where to put the hydrogens. Perhaps a lookup table for one atom is easiest, while for two atoms, Hs are placed along the bond axis, in a plane or as methyl arrangement. For three atoms, I'm not so sure how to solve this.

[Dom1L]

For three atoms you could also just place them along the bond axis followed by a minimization using the lsbuild error function to optimize w.r.t. the distances