Prot42 : a Novel Family of Protein Language Models for Target-aware Protein Binder Generation

Current protein engineering methods are expensive and time-consuming. While recent AI approaches like AlphaProteo and RFdiffusion can design protein binders, they require detailed 3D structures and binding site information that's often unavailable. We present Prot42, a family of protein language models trained on large protein sequence databases that can generate binders from sequence information alone. Unlike existing models limited to ~1,000 amino acids, Prot42 handles sequences up to 8,192 residues, enabling it to work with large proteins and multi-domain complexes. We demonstrate that Prot42 can generate high-affinity protein binders , offering a practical sequence-only approach to computational protein design.