Abstract
Competition-based immunoassays are a common strategy for detecting small-molecule biomarkers. However, these assays rely on the availability of a custom competitor molecule that can effectively be displaced upon analyte binding, often requiring time-consuming synthesis and conjugation steps. De novo designed protein binders present a compelling alternative, as their binding properties can be tuned and they allow for straightforward genetic-incorporation into existing immunoassays. Here, we leverage the BindCraft pipeline to design competitive binders by targeting antigen-binding sites, followed by in silico filtering to select for steric clashes with the small-molecule analyte. As a proof of concept, we designed digoxin competitors and experimentally screened the binders using a simple bioluminescent assay, identifying 7/10 successful binders directly in bacterial lysate. These binders exhibited low to moderate binding affinities (K (d) = 42 nM - 1.1 μM) and were displaced by digoxin. Two de novo binders were encoded into a previously established competition-based immunosensor, enabling sensitive digoxin detection (K (d) = 109 nM). These results demonstrate that deep learning-based models can rapidly yield effective competitor binders, enabling straightforward adaptation and optimization of sensing platforms for small-molecule targets.