Fully human heavy-chain-only antibodies (HCAbs) generate single-domain building blocks that do not occur anywhere in nature on their own. Nona Biosciences’ Hu-mAtrIx™ (Nona’s AI platform for antibody lead selection and developability optimization) integrates directly with the fully human HCAb VH domains produced by Harbour Mice® (transgenic mice engineered to produce fully human heavy-chain-only antibodies) to guide sequence selection toward higher-affinity, higher-developability candidates. The following questions address how this AI-guided discovery process works, what makes it different from generic “AI-assisted” antibody design, and where it fits into a preclinical development timeline.
What does “de novo VH domain discovery” actually mean?
Autonomous, fully human VH domains do not exist naturally as standalone proteins. VH domains may be sourced from synthetic phage display libraries; however, extensive modifications are required to improve upon their solubility and stability. In Nona’s discovery process, HCAb VH domains are instead generated in vivo through natural immune maturation in Harbour Mice®, then further refined computationally, an approach distinct from purely synthetic library engineering that requires heavy downstream re-engineering to reach a usable therapeutic candidate.
How is a Harbour Mice® HCAb VH single domain different from a camelid VHH?
A Harbour Mice® HCAb VH single domain is produced entirely from human VH gene segments in a transgenic mouse, while a VHH (nanobody) originates from camelid immune systems and carries non-human sequence. Engineering efforts on synthetic VH domains have focused on reducing hydrophobicity within the VH/VL interface (FR2) and introducing amino acid substitutions into the CDR sequences, a camelization process that represents a double-edged sword, in that changes introduced can end up promoting aggregation, forming new epitopes that increase immunogenicity, or reducing antigen binding specificity and affinity. By contrast, VH domains sourced from transgenic mice, such as the HCAb Harbour Mice® engineered with fully human VH genes, undergo a natural process of selection, with naturally occurring amino acid substitutions that help impart improved solubility properties to the fully human HCAb VHs. This is a key point of terminology confusion in the field: VHH should be reserved exclusively for camelid-derived single domains, while HCAb-derived single domains from Harbour Mice® should always be called VH domains, never VHH, since the two are structurally and immunologically distinct.
Are fully human HCAb VH domains automatically free of immunogenicity risk?
No, fully human sequence origin reduces immunogenicity risk but does not eliminate it automatically. As fully human sequences, it would be easy to assume that this approach will eliminate immunogenicity risks. Yet, a Phase I study evaluating the safety, tolerability, and pharmacokinetics of GSK1995057, a fully human, single-domain VH antibody directed against the TNFR1 receptor, was halted due to unexpected immunogenicity and the development of cytokine release syndrome, with some affected subjects found to have preexisting antibodies targeting the molecule. The underlying cause was traceable to a specific structural liability rather than the fully human origin itself: based on structural analysis, immunogenicity was largely linked to a proline residue found near the C-terminus, and the ease of engineering VH binders allowed a rapid pivot by adding a single alanine residue to the C-terminus sequence, reducing the frequency of anti-drug antibodies from 50% to 15%. This case illustrates exactly why AI-guided developability screening matters: catching this kind of liability computationally before it reaches a Phase I trial protects both timeline and patient safety.
How does Hu-mAtrIx™ differ from generic “AI-assisted” antibody design?
Hu-mAtrIx™ is an end-to-end integration of AI capabilities with Nona’s antibody discovery and engineering expertise, not a bolt-on prediction tool applied to an otherwise conventional workflow. Most platforms marketed as “AI-assisted” apply machine learning to a single narrow step, such as sequence scoring after candidates are already selected. Nona Biosciences’ Hu-mAtrIx™ AI-platform integrated in discovery extends this further by guiding the incorporation of developability-optimized sequences from the earliest stages of lead generation, rather than filtering candidates only after they have already been synthesized and tested.
What specific developability problems does AI screening catch that traditional screening misses?
AI-based developability scoring identifies structural liabilities such as poor folding or aggregation risk before a candidate is ever synthesized in the lab. Traditional screening evaluates these attributes step by step through a funneled process, and it is common under that model to discover late that an otherwise potent candidate has low yield due to folding or aggregation defects. Applying computational developability assessment at the earliest triage stage allows these risks to be identified and avoided from the outset, improving the overall success rate of a discovery campaign rather than deferring the failure to a costlier, later stage.
At what point in a discovery program does AI developability scoring matter most?
Early lead triage is the decision point where AI developability scoring delivers the greatest impact. This is the stage at which a team must choose which subset of binders advances to synthesis and testing, and choosing well determines whether the program carries forward a diverse set of sequences free of manufacturability risk. Getting this decision wrong means investing downstream resources into candidates that were always going to fail on solubility, aggregation, or expression grounds, so front-loading the developability assessment protects both budget and timeline.
Can HCAb VH domains match the affinity of conventional IgG antibodies?
Yes, HCAbs routinely meet or exceed the binding strength of conventional IgG antibodies despite lacking a light chain. A common concern is whether HCAbs, which lack light chains, can achieve affinities comparable to conventional IgG antibodies. The Harbour Mice® platforms are engineered with fully human VH repertoires, enabling affinity maturation through natural and robust in vivo immune processes that allow for natural somatic hypermutation. Nona’s immunization protocols and proprietary screening technologies routinely yield HCAbs with sub-nanomolar to low nanomolar affinities. The single-domain nature of VHs also allows for streamlined and efficient downstream affinity optimization via standard molecular engineering techniques if further refinement is required.
Where does AI still fall short in antibody discovery, and where is human judgment still required?
Data scarcity remains the central limiting factor for applying AI broadly across antibody discovery. Novel drug development carries inherent risk, and human scientists must weigh multi-dimensional information that extends beyond the antibody itself, including target biology, experimental reality, and the competitive landscape. AI models are trained on limited and incomplete historical program data, which makes generalization to novel targets or entirely new modalities extremely difficult without expert oversight guiding the interpretation.
How does a fully human VH domain compare structurally to an scFv or a VHH?
The comparison comes down to three variables: size, human sequence content, and re-engineering burden.
|
Format |
Origin |
Key Limitation |
|---|---|---|
|
scFv |
Human/humanized V regions linked synthetically |
Requires extensive engineering to overcome stability and immunogenicity issues from linker design |
|
VHH |
Camelid (llama/camel) |
Favorable biophysics but requires humanization and may still trigger immune responses |
|
HCAb VH domain (Harbour Mice®) |
Fully human, in vivo selected |
Requires developability screening but starts from a human sequence baseline |
Fully human variable heavy chain domains sourced from heavy-chain-only antibodies combine the solubility and size advantages of VHHs with the human origin needed for clinical development. This directly fulfills the need of investigators for compact VH binders that combine the developability advantages of VHHs with the clinical readiness of human sequences, eliminating the pitfalls of camelid humanization or scFv linker engineering.
Why haven’t fully human VH domains reached approval yet if the format has these advantages?
Adoption lags because scFv and VHH platforms have a longer track record of validated engineering and manufacturing infrastructure. In the current clinical landscape, more approved biotherapeutics are based on scFv over VHH binders, and more molecules in clinical development utilize scFvs (approximately 36%) than VHH (approximately 10%) binders. To date, no approved biotherapeutics have leveraged fully human VH domains, although a number of candidates have entered clinical evaluation. This trend is partly motivated by the legacy of established engineering platforms, scalable expression systems, and validated clinical precedents rather than any inherent structural disadvantage of the fully human VH format itself. Developers evaluating a bispecific or multispecific program can review Nona’s approach to bispecific and multispecific antibody engineering to see how HCAb VH domains are being applied to next-generation constructs, or explore the underlying HCAb platform technology directly.
Partnering with Nona provides access to AI-guided HCAb VH domain discovery that combines two decades of Harbour Mice® optimization with Hu-mAtrIx™ developability scoring at the earliest stages of candidate selection, reducing the re-engineering cycles that have historically slowed fully human VH programs toward IND filing. Developers ready to explore how this approach fits a specific target or modality can connect with Nona’s discovery team through the Idea toward IND (I-to-I®) framework to scope a program from early discovery through IND-enabling studies.
-
Rouet R, et al. Fully Human VH Single Domains. J Biol Chem, 2015. Link
-
Bélanger K, et al. Optimization of synthetic human VH affinity and solubility. Protein Sci, 2025. Link
-
Gan X, et al. An anti-CTLA-4 heavy chain-only antibody with enhanced Treg depletion. PNAS, 2022. Link
-
Proudfoot A, et al. Novel anti-tumour necrosis factor receptor-1 (TNFR1) domain antibody. Thorax, 2018. Link
-
Mullin M, et al. Applications and challenges in designing VHH-based bispecific antibodies. MAbs, 2024. Link
-
Cordy JC, et al. Specificity of human anti-variable heavy (VH) chain autoantibodies. Clin Exp Immunol, 2015. Link
-
Papadopoulos KP, et al. Unexpected hepatotoxicity in a phase I study of TAS266. Cancer Chemother Pharmacol, 2015. Link
-
Gil D, Schrum A. Strategies to stabilize compact folding. Adv Biosci Biotechnol, 2013. Link
