HKeyBio Industry Commentary

Preclinical Models for Predicting ADA Risk

Translational Medicine Series 07

Originally published in Chinese on September 17, 2026. This commentary reflects information available as of September 8, 2026.

HKeyBio’s take. Can the absence of detectable anti-drug antibodies (ADAs) in an animal study establish a low risk of human immunogenicity? Can a favorable computational score replace experiments with human immune cells? The current evidence supports a more qualified conclusion: preclinical tools help identify risks, compare candidates, and guide molecular optimization, but they cannot yet predict the incidence of ADAs in patients with general reliability.

This edition of HKeyBio’s Translational Medicine Series focuses on therapeutic proteins, particularly the risk of anti-drug antibodies (ADAs) to antibodies and fusion proteins. These conclusions should not be directly extrapolated to vaccines, cell therapies, or gene therapies. Drawing on regulatory documents and research available for review as of September 8, 2026, it examines the current models and their translational boundaries.

Summary. Reading guide

01 Define the prediction target and interpret model results

02 Tools resolve aspects of risk but have different blind spots

03 Animal and emerging models each have appropriate applications

04 Use multiple layers of evidence for decisions rather than a composite score

01 Define the Prediction Target and Interpret Model Results

ADA formation is one outcome of immunogenicity, not its entirety. In a typical T-cell-dependent antibody response, a therapeutic protein is taken up, processed, and presented as peptides; helper T-cell responses may subsequently support B-cell differentiation and antibody production. An individual assay usually captures only selected steps. A positive result at one stage does not mean the full response will necessarily occur. [1,3]

Three distinct outcomes must be considered: whether ADAs are generated, whether they possess neutralizing activity, and whether they affect drug exposure, safety, or efficacy. Non-neutralizing ADAs can still alter pharmacokinetics (PK), while ADA detection alone does not establish clinical harm. The first question when interpreting a model is therefore whether it measures antigen presentation, cellular activation, or antibodies that have actually formed. [1,2]

 Figure 1. General mechanism of T-cell-dependent ADA formation. Reproduced from Karle et al. [3], Figure 1, under CC BY 4.0. © 2025 Karle et al. Created in BioRender. Karle, A. (2025), https://BioRender.com/z1efgeo.

02 Tools Resolve Aspects of Risk but Have Different Blind Spots

Computational Prediction Supports Early Screening but Scores Are Not Incidence Rates

Methods such as NetMHCIIpan estimate peptide binding to, or presentation by, human leukocyte antigen class II (HLA-II) molecules. Version 4.3 incorporates information relevant to HLA-DR, -DQ, and -DP and can help identify candidate sequence regions for further examination [4]. A ranking score, however, neither proves T-cell recognition nor represents a patient-level probability of ADA development. Analyses should document the HLA alleles represented in the target population and the software version, ensuring that findings are interpretable and reproducible.

MAPPs Reveals Presentation but Functional Validation Remains Necessary

Major histocompatibility complex-associated peptide proteomics (MAPPs) typically exposes human antigen-presenting cells to a drug candidate, then uses liquid chromatography–tandem mass spectrometry to identify HLA-II-associated peptides. This measures naturally processed and presented peptides, providing useful evidence for epitope mapping and molecule-to-molecule comparisons. Nevertheless, peptide detection does not demonstrate subsequent T-cell activation, and non-detection can reflect donor selection or the depth of analytical coverage. [3]

Human T Cell Assays Assess Function but Can Miss Responses

Peripheral blood mononuclear cell (PBMC) systems and dendritic-cell/autologous CD4+ T-cell co-cultures can measure functional responses, including proliferation. In a 2024 study of ten antibodies, a CD8-depleted PBMC assay detected responses to four of eight antibodies with relatively high clinical immunogenicity, while a dendritic-cell/CD4+ T-cell co-culture detected five; three were negative in both systems [5]. These findings illustrate why a single T-cell assay cannot predict the clinical immunogenicity of every biotherapeutic.

Best-practice recommendations published in January 2026 emphasize donor selection, cell quality, controls, and predefined interpretation criteria [6]. For drug development programs in China, donor number should be considered alongside how well the HLA distribution represents the intended patient population. Molecules that directly alter immune-cell function need especially careful interpretation: a low assay response caused by pharmacological suppression should not automatically be classified as low immunogenicity risk.

Table 1. Uses and major limitations of preclinical assessment tools.

03 Animal and Emerging Models Each Have Appropriate Applications

Animal and Emerging Models Add Value by Addressing Specific Questions

ADA in conventional animals is first a variable that helps explain findings within that study. The European Medicines Agency (EMA) notes that human or humanized proteins can be recognized as foreign by animals and that animal immunogenicity is generally poorly predictive of immunogenicity in humans. ADA measurements in animals may nevertheless help interpret repeat-dose toxicity findings [2]. Thus, ADA positivity in cynomolgus monkeys does not by itself establish high human risk, and an ADA-negative animal study is not proof of human safety.

For preclinical in vivo efficacy studies, a more useful approach is to follow ADA over time alongside changes in drug exposure and efficacy. If efficacy diminishes after repeat dosing, first check whether exposure has fallen and whether drug interference affects the assay, before considering a loss of effect related to the target mechanism. A single positive ADA result at the end of a study is usually insufficient to establish that ADA caused the efficacy change. This is a proposed approach to study interpretation based on the regulatory principles above. [2]

Immune-humanized mice are progressing toward more complete antibody responses. A proof-of-concept study published in August 2026 used human thymic tissue and hematopoietic stem cells in six experimental groups to investigate responses to infliximab and interferon beta-1b. Drug-specific ADAs, including some neutralizing responses, were observed [7]. However, six experimental groups are not equivalent to six fully independent donors, and donor variability and nonspecific responses in some control animals were reported. The results support further development for risk identification; they do not demonstrate reliable ranking across diverse candidates or general regulatory acceptance.

Organoids aim to retain more of the functionality of human immune tissues. A 2021 human tonsil organoid study reproduced antigen-specific antibody responses, class switching, and affinity maturation following stimulation with antigens, including vaccines [8]. This provides a research foundation for studying cooperation between T and B cells, but the study did not validate prediction of ADA incidence for therapeutic proteins. Translating vaccine-response capability into drug-risk prediction still requires drug panels, positive and negative controls, and independent validation.

For both humanized mice and organoids, “more human-like” should specify which immune functions the model reproduces. Donor coverage, the quality of immune reconstitution, antibody class switching, background responses, and reproducibility should be evaluated against the question being asked. The “humanized” label alone does not establish the relative merits of a model.

04 Use Multiple Layers of Evidence for Decisions Rather Than a Composite Score

During discovery and optimization, sequence and product-quality risks can be reviewed first, with computational findings used to identify regions requiring validation. Aggregation, impurities, route of administration, and the intended patient population should also be considered. Reducing a sequence-derived score should not come at the expense of evaluating the product as a whole. [1]

For candidate comparisons, formulation status, donor conditions, controls, and interpretation criteria should be kept as consistent as possible. If MAPPs identifies a newly presented region while a T-cell assay is negative, first examine the biological steps covered by each method and the experimental conditions, then decide what further validation is needed. Two negative results should not be treated as “cancelling out” one positive result. Even minor sequence optimization should be accompanied by confirmation that the original activity and physicochemical quality are retained.

Once in vivo work begins, ADA measurements should be evaluated with exposure, efficacy, and safety readouts. Clinical ADA assay strategy should also be planned before entering the clinic. The US FDA 2019 guidance differentiates screening, confirmatory, titer, and neutralizing antibody assessments and discusses analytical method performance [9]. Risk-prediction models and ADA detection assays address different questions. The latter measure antibodies in samples; establishing a detection assay does not establish the ability to predict human immunogenicity.

When clinical ADA datasets are used to compare molecules or develop models, differences in patient populations, concomitant medication, sampling schedules, and assay methods must be examined. A 2026 analysis combining published in vitro readouts with clinical ADA frequencies explored relationships between mechanistic measurements and clinical outcomes [10]. Retrospective associations may guide method development but do not replace prospective validation for a new candidate.

The practical role of preclinical ADA prediction is to identify risks that merit attention early, explain inconsistent evidence, and support evidence-based changes to molecules or study designs. At present, the most valuable deliverable is often a risk assessment that clearly states its supporting evidence, scope of applicability, and remaining uncertainties.

HKeyBio Perspective

HKeyBio specializes in autoimmune disease models. In preclinical in vivo efficacy evaluation for autoimmune and allergy therapies, we focus on how models, pathology, immune readouts, and analytical testing inform one another. For repeat-dose programs potentially affected by ADA, study design can be discussed in light of the need to interpret changes in exposure and efficacy. Specific models, assays, cross-species suitability, and implementation requirements should be confirmed individually according to the molecule’s characteristics and development stage.

References

[1] FDA. Immunogenicity Assessment for Therapeutic Protein Products. Guidance for Industry. 2014. Source

[2] EMA. Guideline on immunogenicity assessment of therapeutic proteins. Revision 1. 2017. Section 6: non-clinical immunogenicity assessment. Source

[3] Karle AC, et al. MAPPs assays for non-clinical immunogenicity risk assessment: best practices recommended by the European immunogenicity platform. Front Immunol. 2025;16:1690101. doi:10.3389/fimmu.2025.1690101. Figure 1: CC BY 4.0. Source

[4] Nilsson JB, et al. Accurate prediction of HLA class II antigen presentation across all loci using tailored data acquisition and refined machine learning. Sci Adv. 2023. doi:10.1126/sciadv.adj6367. See also DTU documentation for software versions. Source

[5] Walsh RE, et al. Preclinical immunogenicity risk assessment of biotherapeutics using CD4 T cell assays. Front Immunol. 2024;15:1406040. doi:10.3389/fimmu.2024.1406040. Source

[6] Tourdot S, et al. T cell assays for non-clinical immunogenicity risk assessment: best practices recommended by the European Immunogenicity Platform. Front Immunol. 2025;16:1723110. Published January 12, 2026. doi:10.3389/fimmu.2025.1723110. Source

[7] Gharaie S, et al. NeoThy™ immune-humanized mice can produce anti-drug antibodies to support immunogenicity assessment for biological drug products. Front Immunol. 2026;17:1851898. doi:10.3389/fimmu.2026.1851898. Source

[8] Wagar LE, et al. Modeling human adaptive immune responses with tonsil organoids. Nat Med. 2021. doi:10.1038/s41591-020-01145-0. Source

[9] FDA. Immunogenicity Testing of Therapeutic Protein Products—Developing and Validating Assays for Anti-Drug Antibody Detection. Guidance for Industry. 2019. Source

[10] Agnihotri S, Venkatesh A, Ramanathan M, Balu-Iyer S. First step towards predicting clinical immunogenicity of biologics using in vitro based readouts as animal trial alternatives. J Pharm Sci. 2026;115(6):104297. doi:10.1016/j.xphs.2026.104297. Source

NetMHCIIpan 4.3 official documentation. Source

Author: HKeyBio Marketing Department | Scientific review: Jing Yan | Editor: Wang Shuyan

About HKeyBio

HKeyBio, the autoimmune disease model specialist, is a preclinical contract research organization (CRO) focused on autoimmune and allergic diseases, providing end-to-end in vivo efficacy evaluation services. Its core technical team has more than 20 years of experience across the autoimmune drug development process. The company is ISO 9001 certified and operates an integrated research facility in Suzhou Industrial Park covering molecular and cellular research, pathology, small-animal studies, and analytical testing, alongside two nonhuman primate research sites in Suzhou and Guangxi. HKeyBio has nearly 80 employees, and all project leaders hold master’s or doctoral degrees. The team has accumulated experience supporting more than 500 investigational new drug (IND) applications in autoimmune diseases, enabling it to efficiently address complex challenges throughout autoimmune drug development and regulatory submission.

· 50+ nonhuman primate models of autoimmune and allergic diseases

· 500+ animal models of autoimmune and allergic diseases

· Experience supporting 500+ IND applications in autoimmune diseases