Enables zero-shot, directed protein generation by applying a simple scalar bias to stochastic attention samplers.
March 23, 2026
Original Paper
Conditioning Protein Generation via Hopfield Pattern Multiplicity
arXiv · 2603.20115
The Takeaway
Allows practitioners to steer protein generation toward specific functional subsets (like binding or stability) without any model retraining. It identifies a 'calibration gap' based on the geometric separation of functional groups in the encoding space.
From the abstract
Protein sequence generation via stochastic attention produces plausible family members from small alignments without training, but treats all stored sequences equally and cannot direct generation toward a functional subset of interest. We show that a single scalar parameter, added as a bias to the sampler's attention logits, continuously shifts generation from the full family toward a user-specified subset, with no retraining and no change to the model architecture. A practitioner supplies a sma