Interface to AIMNet2

[peastman]

AIMNet2 looks like a good candidate for another model to include. It supports 14 elements and charged molecules. It includes an explicit Coulomb term, which should help with accuracy on large molecules. The current implementation has O(N^2) time and memory in the number of atoms. They say a linear scaling version is coming soon.

[jchodera]

Agreed. Can I put you in touch with the authors?

Do we intend for this to be useful in ML/MM hybrid simulations? Would we envision the ML model to only be used for intra-ML-region interactions, or would we want some part of the model (eg charges) to be used in ML-MM interactions?

[giadefa]

We are already using it for MM/MM and it's good.

…

On Tue, Oct 24, 2023 at 10:02 PM John Chodera ***@***.***> wrote: Agreed. Can I put you in touch with the authors? Do we intend for this to be useful in ML/MM hybrid simulations? Would we envision the ML model to only be used for intra-ML-region interactions, or would we want some part of the model (eg charges) to be used in ML-MM interactions? — Reply to this email directly, view it on GitHub <#63 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3KUOTWSIIDSGV75UOXYELYBANFFAVCNFSM6AAAAAA6OEKQXWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONZXHE2TANRQGQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[peastman]

In principle it should be possible to use the charges it computes to compute interactions between the ML and MM regions. But the first step is to use fixed charges the same way we do with ANI.

[peastman]

Can I put you in touch with the authors?

That would be great.

[RaulPPelaez]

AFAICT they only provide TorchScript compiled files https://github.com/isayevlab/AIMNet2/tree/main/models
They do not provide source code for their models besides these checkpoints in this repo.
TorchForce should be able to handle it just fine.