Scalable De Novo Binder Design Enabled by Integrated Computational Design and High-Throughput Screening

Abstract

Proteins mediate nearly all cellular processes, and selectively binding protein surfaces is central to both biological discovery and therapeutic development. While recent advances in computational de novo design and structure prediction have accelerated in silico modeling, a persistent bottleneck is that computational confidence does not reliably translate into functional binding, especially for short, flexible ligands where conformational dynamics and context-dependent stabilization are difficult to capture with standard metrics. This dissertation develops integrated computational–experimental workflows for de novo discovery and optimization of protein-binding ligands across two complementary modalities. First, it establishes a benchmark for structure-guided design using compact, stable miniprotein binders against Francisella-like lipoprotein 3 (Flpp3), a previously ligand-free protein target, combining large-scale design, yeast surface display screening, and biochemical and structural validation to connect designed models to experimentally realized binding modes. This workflow yielded multiple nanomolar-affinity binders, including three picomolar-affinity leads. An X-ray crystal structure of Flpp3 in complex with a designed binder closely matches the design model (Cα RMSD: 0.9 Å), validating the design model at near-atomic resolution. The work then focuses on disulfide-stapled macrocyclic peptides, a modality that offers compactness and interface adaptability while remaining challenging to design predictively. To enable library-scale evaluation, the thesis introduces a genetically encodable disulfide-stapled format compatible with yeast surface display and optimizes the display architecture to sensitively measure expression and binding for short peptides. To assess generality across diverse binding contexts, this platform was applied to four representative targets -DnaN, GABARAP, PCSK9, and MCL1- collectively evaluating tens of thousands of unique de novo designs and additional mutational variants through high-throughput screening and affinity maturation. These campaigns yielded medium- to high-affinity peptide binders for each target. Together, these studies couple generative design to large-scale experimental selection, clarifying where functional binders emerge relative to computational metrics and enabling iterative, feedback-informed improvements to de novo peptide design in future discovery efforts.