ABLE-Guided Rational Design of High-Avidity Biparatopic Nanobodies

Abstract

High-affinity protein binders are essential tools for molecular diagnostics, imaging, and targeted therapeutics. However, rationally designing nanobodies with ultra-slow dissociation kinetics remains a persistent challenge. While biparatopic nanobody formats can enhance functional avidity through intramolecular rebinding, their development has largely depended on empirical screening rather than predictive design principles.

Here, we present ABLE, a gradient boosting-based model that transforms biparatopic design into a predictive discipline by quantitatively linking structure to function. Using physically interpretable and tunable geometric constraints derived from AlphaFold-predicted antigen-nanobody complexes—contact Area ratio, Bond number ratio, Linker adequacy ratio, and inter-Epitope distance—ABLE prioritizes pairings that maximize cooperative binding. Across viral (SARS-CoV-2 S1 and nucleocapsid proteins), human (cardiac troponin I, cTnI), and bacterial (Mycobacterium tuberculosis Rv2074) targets, the model delivered >400-fold avidity gains in 4.35–17.65% of constructs starting from modest (μM–nM) affinity monomers, with a maximum improvement of 4,714-fold and generating constructs with picomolar apparent avidities.

Applied to the tuberculosis antigen Rv2074, ABLE-designed biparatopic vNARs reached sub-100-pM apparent avidity (61–99 pM) and enabled rapid development of a sandwich chemiluminescent immunoassay. From a 25 × 42 capture-detection matrix screen, the lead combination (BBsR26–16) maintained robust sensitivity across complex biomimetic matrices, achieving a detection limit of 30 pg mL⁻¹ directly in patient urine with measured concentrations (30–299 pg mL⁻¹) closely following the calibration curve (R² = 0.98).

Iterative refinement through a Design-Build-Test-Learn cycle incorporating published and experimental constructs improved predictive performance, with the final model achieving R² = 0.92 on training and 0.84 on held-out test sets, along with 90.50% 2-fold accuracy and 100% 5-fold accuracy. Feature importance analysis confirmed that inter-epitope distance and monomer affinity balance are the dominant determinants of avidity gain, consistent with ABLE's core premise that productive cooperativity requires both geometric compatibility and balanced interfacial contributions.

In summary, ABLE establishes a generalizable, geometry-aware framework that shifts biparatopic nanobody development from empirical screening to rational, structure-based design. By leveraging AlphaFold-predicted structures and modest-affinity monomers, the model achieves high avidity designs without requiring large immune libraries, extensive mutagenesis, or deep learning-based optimization, thereby streamlining development. This work provides a scalable and efficient strategy for creating high-performance constructs, unlocking new opportunities for sensitive diagnostics and potential therapeutic development.

Date of Award	1 Apr 2026
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Xin DENG (Supervisor), Thi Nguyet Minh LE (Supervisor) & Jiahai SHI (Co-supervisor)