Pallatom2: Navigating the De Novo Protein Universe with Atom-to-Atom Modeling

Pallatom2, an atom-to-atom generative binder model, identifies 112 active designs among 210 candidates (53.3%) and yields binders for all ten experimentally evaluated targets.

• Why it matters: Protein binder design depends on coordinating sequence, structure, and interface geometry, but existing models often struggle to balance atomic precision with controllable, target-specific performance.
• What they did: Pallatom2 uses a unified Atom14 representation with residue-guided sparse all-atom attention to model sequence, backbone and side-chain geometry, and target-interface contacts, then screens nine targets in silico.
• The result: It achieves picomolar-scale apparent affinities on six targets (e.g., K_D 3.96 pM for SC2RBD and 33.8 pM for PD-L1) and supports interface transfer, affinity maturation, and recovery of inactive designs.

AI summary of the abstract · may contain errors · not medical advice · read the paper.