A Mammalian High-Throughput Assay to Screen AI-Designed Protein Degraders

Abstract

The development of specific protein binders is crucial for biologics and targeted protein degradation (TPD) therapies. However, existing screening methods are low-throughput, labor-intensive, and often rely on non-human display systems such as phage, yeast, or mRNA display, limiting their translational relevance. To address this, we developed a high-throughput, human cell-based binder screening platform that enables the functional evaluation of artificial intelligence (AI)-designed peptide binders in a mammalian context. Our approach utilizes genetically-encodable, doxycycline-inducible ubiquibodies (uAbs), where a library of computationally designed ``guide'' peptides is fused to an E3 ubiquitin ligase domain, enabling modular, CRISPR-like TPD. By monitoring degradation through an mCherry-fused target protein, we screen and validate AI-generated binders in a physiologically relevant setting. We successfully apply this platform to identify functional binders for EWS::FLI1 and $\beta$-catenin, two highly challenging oncogenic targets. Overall, our approach facilitates the discovery of AI-designed binders for diverse therapeutic applications.