Pallatom2: a 65.9M-param atom-level protein design model hits 53.3% experimental success rate

sokrypton · x · 2026-11-09

Hangzhou-based Levinthal Biotech introduced Pallatom2, an atom-to-atom generative model for de novo protein binder design.

Method

A unified Atom14 representation merges backbone, side chains, and L/D residues;
AtomPairFormer encodes 3D geometry and bonds;
Residue-guided sparse all-atom attention links global structure to local atomic detail.

Key results

Only 65.9M parameters, yet monomer sampling extends to 1,600 residues across monomers, multi-domain proteins, and homomultimers;
Binder design: 112 of 210 tested candidates were active (53.3%), with binders found for all 10 experimentally evaluated targets;
Six targets reached picomolar apparent affinities, including KD 3.96 pM for SC2RBD and 33.8 pM for PD-L1;
Partial diffusion improves TNFα affinity up to 7.57-fold, and CDR-associated interface remodeling converts an inactive de novo nanobody into binders with KD of 4.31/5.93 nM.

More from Apps