An AI-driven approach to nanobody engineering is drawing federal support, with Seattle Children's Hospital receiving a USD 2.1 million NIH R01 grant to develop a platform for designing intracellularly functional, conditionally stable nanobodies — a class of single-domain antibody fragments that has long been constrained by off-target accumulation inside living cells.
The four-year award, funded by the National Institute of General Medical Sciences, runs through April 2030 and is led by principal investigator Jonathan C. Y. Tang, whose prior work spans Columbia University, the University of Washington, and the Allen Institute.
The intracellular nanobody problem
Nanobodies — camelid-derived, single-domain antibody fragments — are genetically encodable and small enough to function inside cells, making them attractive as research tools and potential therapeutic agents. The core limitation is that most nanobodies are constitutively stable, meaning they accumulate regardless of whether they have engaged their target. That background signal degrades specificity and limits their utility for low-noise intracellular detection or regulated protein modulation.
Tang's group is addressing this by engineering what the grant terms "conditionally stable" nanobodies, or CS-NBs: variants modified by AI to be inherently unstable, but stabilized only upon binding their intended target. The design logic reduces background accumulation and improves target specificity without requiring exogenous small-molecule switches or complex delivery systems.
Platform design and validation scope
The project integrates deep learning, structural bioinformatics, and high-throughput cell-based screening across three stated aims. The first expands an existing AI-based stability prediction tool using data from more than 2,000 nanobody-reporter fusions assayed in human 293T cells via high-content microscopy — a dataset scale that should meaningfully extend the model's ability to discriminate stable from unstable intracellular variants.
The second aim systematically maps and mutates non-interface nanobody residues across structural classes to define sequence-level rules governing conditional stability without compromising target affinity. Those findings feed back into the predictive model, generating what the investigators describe as a generalizable "destabilizing switch" applicable to newly discovered nanobodies.