Wrote the below before remembering “Jude you dumb idiot, PAHs are stupid insoluble, that’s like… that’s going to stop you no matter what enzyme you make. I guess you could try to do it in organic solvent, but that’s another huge wrinkle at a minimum.” Anyways, the below is cool news and still stands for things less hellish than graphene.
Actually, it turns out that “near future” was “negative one day” in this case. DISCO was just released, an AF3 successor model for sequence and structure co-design. At a very quick glance (haven’t dug deep into it yet), it generated enzymes for some new-to-nature reactions by conditioning on “bind-to-the-transition-state-geometry, sorta.”
So if the Diels-Alder thermodynamics check out, and the gippities are optimistic but I haven’t verified it yet, I retract my prior statement. Someone could reasonably try this as a project.
You could probably just condition jointly on “is a functional DAase for a bunch of large polycyclic aromatic hydrocarbons.” Who knows if that’d yield a functional enzyme with DISCO, but if it doesn’t, the models will probably be in the right place not too long from now.
EDIT: Friend who works directly on the problem says the paper is very transparent, but more wet lab is needed to see how far these things generalize. Regardless, a good proof of principle.
Wrote the below before remembering “Jude you dumb idiot, PAHs are stupid insoluble, that’s like… that’s going to stop you no matter what enzyme you make. I guess you could try to do it in organic solvent, but that’s another huge wrinkle at a minimum.” Anyways, the below is cool news and still stands for things less hellish than graphene.
Actually, it turns out that “near future” was “negative one day” in this case. DISCO was just released, an AF3 successor model for sequence and structure co-design. At a very quick glance (haven’t dug deep into it yet), it generated enzymes for some new-to-nature reactions by conditioning on “bind-to-the-transition-state-geometry, sorta.”
So if the Diels-Alder thermodynamics check out, and the gippities are optimistic but I haven’t verified it yet, I retract my prior statement. Someone could reasonably try this as a project.
You could probably just condition jointly on “is a functional DAase for a bunch of large polycyclic aromatic hydrocarbons.” Who knows if that’d yield a functional enzyme with DISCO, but if it doesn’t, the models will probably be in the right place not too long from now.
EDIT: Friend who works directly on the problem says the paper is very transparent, but more wet lab is needed to see how far these things generalize. Regardless, a good proof of principle.