Artificial transmembrane transporters
"Generative Landscapes and Dynamics to Design Functional Multidomain Artificial Transmembrane Transporters" — ACS Central Science, July 2025. Two artificial ATP-driven copper transporters, designed from evolutionary sequence constraints rather than copied from nature, and shown to move metal with native-like function.
What it does
- Designed multidomain transporters with a latent generative landscape trained on evolutionary sequence constraints, which puts the design in sequence space no natural protein occupies.
- Paired the generative step with molecular dynamics, so a candidate had to survive as a moving structure and not only look right as a sequence.
- My part ran from the computational design of a copper-transporting P-type ATPase variant through to the wet-lab side: expression, FPLC purification, and reconstitution into proteoliposomes for transport assays.
- The result that matters is the boring one — the designs transport copper. A sequence a model likes is a hypothesis until something in a tube moves an ion.
Notes
Designing a protein that folds is one problem. Designing one that folds, moves, and does a job across a membrane is a different and much less forgiving one, because the parts interact — a sequence that satisfies the fold can be dead on arrival once it has to hinge and bind and let go in the right order.
What the work argues is that you cannot separate those questions and answer them one at a time. The generative model proposes sequences by learning how evolution has constrained them; the dynamics simulations show whether the proposal holds up as a machine; the assays settle it. I spent most of my time on the last stretch of that pipeline, which is where designs stop being promising and start being either functional or not.