AI-designed nanobodies for detection and manipulation applications inside living cells

The National Institute of General Medical Sciences awarded Seattle Children's Research Institute $2.096 million on May 1, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop an AI-driven platform for generating conditionally stable nanobodies for intracellular detection and protein manipulation applications.

The project spans four years through April 30, 2030, and will deliver three core outputs. First, an expanded AI-based nanobody stability prediction tool trained on high-content microscopy data from over 2,000 nanobody-reporter fusions assayed in human cells to improve the model's ability to distinguish stable versus unstable variants inside living cells. Second, systematic mapping and mutation studies of non-interface nanobody residues across diverse structural classes to define sequence rules for conditional stability without compromising target affinity, feeding refined predictions back into the model. Third, validation in the mouse retina as a robust in vivo system to confirm that conditionally stable nanobodies engineered in cell culture retain low background and target-dependent stabilization under physiological conditions. The broader deliverables include an open-source bioinformatics tool and large annotated datasets of nanobody variants.

Work is performed in Seattle, Washington.

Abstract

Nanobodies (NBS)-small, single-domain antibody fragments derived from camelids-hold exceptional promise as genetically encodable binders for intracellular applications. Despite rapid advances in NB discovery, many NBS are constitutively stable and prone to off-target accumulation, limiting their use for low- background detection or regulated protein modulation inside living cells. This project addresses that GAP by developing an AI-driven, scalable platform to generate "conditionally stable" NBS (CS-NBS) from input NB sequences. Sequences modified to be unstable by AI become stabilized only when bound to their target, thereby minimizing background and improving target specificity. We propose three specific aims. Aim 1 will expand an AI-based NB stability prediction tool using large-scale high-content microscopy data (>2,000 NB-reporter fusions assayed in human 293T cells). This will deepen our model's ability to distinguish stable VS. Unstable NB variants inside living cells, accelerating CS-NB development. Aim 2 systematically maps and mutates non-interface NB residues across diverse structural classes, defining the sequence rules that drive conditional stability without compromising target affinity. These findings will feed back into our predictive model, generating a refined "destabilizing switch" strategy applicable to any newly discovered NB. Aim 3 will validate the platform in a robust in vivo system, the mouse retina, confirming that CS-NBS engineered in cell culture retain low background and target-dependent stabilization under physiological conditions. By integrating deep learning, structural bioinformatics, and high-throughput validation, this project will deliver a scalable pipeline for creating intracellularly functional, low-background NB reagents. The outcomes-an open- source bioinformatics tool, large, annotated datasets of NB variants, and validated CS-NB constructs-will be broadly disseminated (e.g., github, addgene), enabling the research community to harness the rapidly expanding NB landscape. This technology-focused work aligns with nigms's mission to develop innovative, enabling tools that advance fundamental biomedical science and lay groundwork for future diagnostic and therapeutic strategies.