Smarter Lead Selection: How AI Developability Scoring Reduces Late-Stage Attrition

Late-stage attrition remains one of the costliest problems in antibody drug development, often traced back to developability issues that were invisible during early screening. AI developability scoring addresses this by evaluating manufacturability, stability, and expression risk at the lead selection stage rather than after significant time and capital have already been invested. Nona Biosciences applies this approach across its discovery workflow, using AI models to prioritize which candidates advance to wet lab validation.

What is AI developability scoring in antibody discovery?

AI developability scoring is a computational method that predicts manufacturability and stability risks in antibody candidates before they undergo full experimental characterization. Rather than relying solely on binding affinity to rank leads, this approach layers in predictions about aggregation propensity, expression yield, and chemical liabilities. Nona adopts an end-to-end AI integration strategy where AI is embedded in every step of the antibody discovery workflow, from target validation all the way to antibody engineering and CMC, with AI models taking the wheel to generate novel sequences and determine which variants are prioritized in wet lab validation. This differs from using AI as a single-purpose filter bolted onto an otherwise conventional pipeline.

Why does developability scoring matter more at early lead triage than at later stages?

Early lead triage is where the largest number of molecules get eliminated, making it the highest-leverage point for catching risk before it compounds. AI developability scoring is applicable in various stages where decisions are made about which molecules should survive to the next step, and during early lead triage in particular, choosing the right subset of binders to be further synthesized and tested is of great importance to advance a diverse set of sequences that exhibit no developability or manufacturability risks. Catching a poor-folding or aggregation-prone sequence at this stage avoids the sunk cost of carrying it through synthesis, expression, and characterization only to discard it later.

How does AI-based screening prevent the late-stage failures that traditional screening tends to miss?

Traditional screening is a funneled process where developability and manufacturability attributes are assessed step by step, and it is common to encounter potent candidates that are found to have low yield because of poor folding or aggregation. These defects and risks can be effectively identified at the very beginning with AI-based developability assessment and subsequently avoided to improve the success rate of discovery campaigns. This shift moves risk detection from a late, expensive checkpoint to an early, low-cost one.

What specific properties does Nona’s developability assessment pipeline evaluate?

Nona’s integrated developability assessment pipeline evaluates four core categories of manufacturability risk before a candidate advances. Nona Biosciences’ integrated developability assessment pipeline evaluates expression yield, thermal stability, aggregation resistance, and chemical liabilities early in the discovery process. This enables rapid identification of candidates with optimal properties for downstream development, minimizing late-stage attrition. These categories map directly onto the failure modes that most often derail candidates during CMC and formulation work later in a program.

Why do fully human HCAb VH domains score more favorably on developability than conventional IgG formats?

Fully human heavy-chain-only antibodies (HCAbs) carry structural advantages that translate directly into stronger developability scores. HCAbs derived from the Harbour Mice® platform are fully human in sequence, enhancing their compatibility with human immune tolerance, reducing the risk of immunogenicity, and potentially facilitating regulatory approval. Their small size and heavy chain-only conformation, compared to the conventional IgG format at roughly 90 kDa for HCAb IgG versus roughly 150 kDa for H2L2 IgG, facilitate expression in cell systems and eliminate the risk of chain mispairing. Moreover, HCAb monomeric VH domains at roughly 12 to 15 kDa enable high-yield expression in both microbial systems and mammalian cell systems such as CHO cells, streamlining production. This structural simplicity gives AI developability models cleaner signal to work with compared to two heavy and two light chain formats, where mispairing risk adds an additional layer of manufacturability uncertainty. Nona’s platform for evaluating this distinction is detailed on its page dedicated to fully human antibody discovery.

Is a fully human HCAb the same as a humanized antibody or a VHH nanobody?

No, these terms describe distinct molecules with different origins and different immunogenicity risk profiles, and mixing them up is a common source of confusion among buyers. Clients often search for “humanized antibodies” or “VHH” when they actually mean fully human single-domain antibodies. A fully human HCAb VH domain is generated entirely from human sequence through natural immune selection in Harbour Mice® (transgenic mice engineered to produce fully human heavy-chain-only antibodies), while a humanized antibody starts as a non-human sequence that is engineered afterward to reduce immunogenicity and still carries residual non-human residues. VHH refers specifically to camelid-derived nanobodies from llamas or camels, a distinct lineage from Nona’s HCAb VH domains, which originate exclusively from human V-gene segments. Because fully human sequences are inherently compatible with human immune tolerance, they carry a lower baseline immunogenicity risk than humanized sequences, and this distinction feeds directly into how AI developability models score candidates.

How does AI developability scoring compare to Nona’s Hu-mAtrIx™ platform?

AI developability scoring is one input into a broader optimization process, and Hu-mAtrIx™ (Nona’s AI platform for antibody lead selection and developability optimization) is the platform that integrates this scoring into actionable lead selection decisions. Nona Biosciences’ Hu-mAtrIx™ AI platform integrated in discovery extends this further by guiding the incorporation of developability-optimized sequences, rather than treating developability as a downstream filter applied after leads are already locked in. This is a meaningful distinction for buyers evaluating vendors: many CROs apply AI narrowly to sequence prediction or affinity ranking, while Nona embeds it across the full path from lead triage through engineering, as outlined on the antibody engineering capability page.

Where does AI developability scoring still fall short, and where is human judgment still required?

Data scarcity remains the primary limiting factor for applying AI broadly across antibody discovery. Novel drug development is a high-risk process where human scientists have to make judgment based on multi-dimensional information, not only trade-offs on the antibodies but also on the target biology, the experimental reality, and the competitor landscape. AI models are trained on limited and incomplete data from historical programs, and it is extremely difficult to generalize to new situations such as novel targets or new modalities. This is precisely why Nona pairs AI scoring with an experienced discovery team rather than deploying models as a fully autonomous decision layer, and why the platform operates a continuous data flywheel to refine model accuracy over time.

When in a discovery program should a developer prioritize AI developability scoring over standard binding-affinity screening?

Developability scoring should be applied as early as possible, ideally at the lead triage stage before antibodies are generated and tested individually, rather than being reserved for late-stage confirmation. Waiting until CMC to assess manufacturability attributes is one of the most common and avoidable gaps in mid-program partnerships, since discovery-stage molecules are frequently selected purely on affinity and epitope with limited assessment of aggregation propensity, chemical stability, or viscosity behavior. Programs pursuing multispecific or bispecific formats benefit even more from early scoring, since these formats introduce additional manufacturability and species-gap complexity that compounds if left undetected, an area covered in depth on Nona’s bispecific and multispecific engineering page. HCAbs from Harbour Mice® are particularly well suited to this early-scoring approach because their reduced structural complexity, free of chain mispairing risk, gives AI models cleaner data to score against from the outset.

Partnering with Nona offers developers a discovery pathway where AI developability scoring is not an afterthought layered onto binding data, but a decision point embedded from the earliest stage of lead triage through to IND-ready CMC. This integrated approach, delivered through Nona’s Idea toward IND (I-to-I®) framework (Nona’s integrated end-to-end service pathway from ideation through IND filing), is designed to reduce the rework and delay that late-discovered developability issues typically cause. Developers evaluating discovery partners can review Nona’s full developability assessment methodology to see how to select an antibody discovery partner that effectively integrates these computational tools across the workflow.

1. Wang C, Vemulapalli B, Cao M, et al., “A systematic approach for analysis and characterization of mispairing in bispecific antibodies with asymmetric architecture,” mAbs, 2018. Link

2. Jain T. et al., Biophysical properties of the clinical-stage antibody landscape, PNAS, 2017. Link

3. Xu Y. et al., Structure, heterogeneity and developability assessment of therapeutic antibodies, mAbs, 2019. Link

4. Muyldermans S., Nanobodies: natural single-domain antibodies, Annual Review of Biochemistry, 2013. Link

5. Raybould M.I.J. et al., Five computational developability guidelines for therapeutic antibody profiling, PNAS, 2019. Link

6. Jarasch A. et al., Developability assessment during the selection of novel therapeutic antibodies, Journal of Pharmaceutical Sciences, 2015. Link

7. Sormanni P. et al., Rapid and accurate in silico solubility screening of a monoclonal antibody library, Scientific Reports, 2017. Link