openff-toolkit

openff-toolkit copied to clipboard

Describe the bug I was looking into the Gen3 torsions set, and there seems to be some chemical perception issue with Nitrogens in resonance cases, or "[#7X3+1]=" type. Here is a notebook with 59 such molecules where some pass through but some fail angle parameter assignment, or torsion parameter assignment.

Charge assignment failures: 0 Angle assignment failures: 1 Proper torsion assignment failures: 13 Total failures: 14 out of 59

To Reproduce Notebook here: nitrogen_failures.tar.gz Pdf of notebook for easy molecule viewing: check_nitrogen_parameters.pdf

Making a note that new parameters with the smarts patterns "[#7X3+1]-[#6X4]" , "[#7X3+1]-[#7X3]" for central bonds in torsions, and "[#8X]=,:[#7X3+1]-[*]" for angle may cover these, until further discussion.

Is this a new issue (finding some molecules that should be assigned parameters, but aren't) or a regression (an old version of the toolkit did assign parameters, but recent versions aren't)?

I think it is a new issue, mostly lack of parameters and not toolkit related.

Do you think this should be on forcefield tracker instead of here?

Wherever is best for you and the fitting team to keep track of reports of missing parameters - if that's here that's not a problem. I just wanted to determine if this was a bug in the code or not.

Just making a note for some failures, since these may not be important chemistries we can try to modify the general parameters to catch them

• for molecules of type (failure for central bond [3,5]) simple solution would be to make t51: [*:1]-[#6X4:2]-[#7X3:3]-[*:4]" more general with the first and third bonds being more flexible [*:1]~[#6X4:2]-[#7X3:3]~[*:4].
• for molecules of type (failure for central bond [8,6]) t138: "[*:1]~[#7X2:2]-[#7X3:3]~[*:4]" to include #7X4 as well in the third position, making it "[*:1]~[#7X2:2]-[#7X3,#7X4:3]~[*:4]"

Closing this since it is not a toolkit issue.

Out of curiosity, what was the issue/where did the issue arise from?