The Wet-Lab Validation Engine Behind AI-Driven Antibody Discovery

Artificial intelligence (AI) is accelerating antibody discovery by enabling researchers to generate novel sequences, predict antibody-antigen interactions, and prioritize candidates computationally. However, a promising in silico design does not necessarily translate into a functional antibody.

Experimental validation remains essential to confirm whether AI-designed antibodies can be expressed, bind their intended targets, and demonstrate suitable developability properties. As computational design becomes increasingly scalable, high-throughput wet-lab testing is needed to keep pace and provide the reliable data required for further optimization.

Why AI-Designed Antibodies Still Need Experimental Validation

AI-driven antibody design uses computational models to predict antibody structures, binding interactions, and other molecular properties. These approaches can significantly narrow the number of candidates requiring experimental screening.

However, antibody-antigen interactions are complex. Protein flexibility, epitope accessibility, and molecular conditions can influence binding in ways that computational models may not fully capture.

A recent study published in Nature Communications illustrates both the potential and limitations of computational antibody design.

Wu et al. (2026) developed tFold System, a structure-driven computational workflow for designing antibodies against specific epitopes. The researchers generated candidates targeting influenza A hemagglutinin, PD-1, PD-L1, and the SARS-CoV-2 receptor-binding domain.¹

After computational design, the antibodies underwent phage-display screening, recombinant expression, and experimental characterization using surface plasmon resonance (SPR) and competition assays.

The researchers identified high-affinity antibodies, including a PD-L1-binding candidate with a dissociation constant (K_D) of 45 pM.

Yet not all computational predictions translated into the expected experimental results. Some antibodies predicted to recognize particular epitopes did not demonstrate the anticipated competition behavior. For influenza A, the highest-affinity candidate appeared to bind outside the intended epitope.

These findings how computational models can predict promising antibody candidates, but experimental testing determines whether they demonstrate the intended molecular properties.

What Does Wet-Lab Validation Measure?

For AI-designed antibodies, experimental validation typically begins with recombinant expression, followed by binding characterization and, where appropriate, early developability assessment.

Recombinant Antibody Expression

The first step generally involves synthesizing the corresponding DNA, cloning it into expression vectors, and producing the antibody in mammalian cells such as CHO or HEK293.

Expression yield provides an early indication of whether the designed antibody can be produced efficiently. Poorly expressing candidates may require further optimization, even if their predicted binding properties are favorable.

Binding and Affinity Characterization

Techniques such as biolayer interferometry (BLI) and SPR can measure antibody–antigen interactions, including binding kinetics and affinity.

These experimental measurements allow researchers to compare predicted binding performance with observed results. Competition assays and other functional tests may be necessary when antibodies are designed to recognize specific epitopes or block biological interactions.

Early Developability Assessment

Besides target binding, candidates must possess suitable properties for downstream development.

Thermal stability, self-interaction, and polyreactivity measurements can help identify antibodies that may present challenges during manufacturing or formulation.

Evaluating these properties early supports candidate selection based on multiple characteristics rather than binding affinity alone.

Scaling Wet-Lab Validation for AI Antibody Discovery

As AI models generate larger numbers of antibody candidates, experimental throughput becomes an important consideration.

Conventional antibody production and characterization often involve separate steps for gene synthesis, expression, purification, and analytical testing. Coordinating these processes for hundreds or thousands of sequences can create bottlenecks.

High-throughput experimental platforms help address this challenge by enabling larger panels of antibodies to be produced and characterized through coordinated workflows.

However, generating more data is only useful when the results are reliable and comparable. Consistent assay conditions, quality controls, and clear links between antibody sequences and measured properties are essential for meaningful analysis.

Both successful and unsuccessful candidates can provide valuable information. For example, an antibody with weak binding or poor expression may reveal limitations in the computational model that generated it.

RushData: High-Throughput Experimental Validation for AI-Designed Antibodies

Biointron developed RushData, an integrated wet-lab platform designed to connect computational antibody sequences with experimental measurements.

Built around Biointron's rapid CHO expression technology, RushData combines gene synthesis, recombinant antibody production, and high-throughput characterization to support AI-driven antibody discovery.

The platform supports 20,000+ antibody molecules per week, helping researchers evaluate large candidate panels.

Depending on project requirements, RushData provides experimental measurements including:

• Antibody expression: Rapid CHO-based production of designed antibody sequences.

• Binding characterization: BLI and/or SPR measurements to evaluate antibody–antigen interactions.

• Developability profiling: Optional thermal stability, self-interaction, and polyreactivity assessments.

• Structured data delivery: Organized experimental datasets linked to antibody candidates for downstream analysis.

By integrating these steps, RushData helps researchers obtain experimental feedback without coordinating multiple disconnected laboratory workflows.

Closing the Loop Between Computational Design and Experimental Data

AI-driven antibody discovery is increasingly moving toward iterative workflows in which experimental results inform subsequent rounds of computational design. This approach follows a Design–Build–Test–Learn cycle:

Computational Design → Antibody Expression → Experimental Testing → Data Analysis → Model Refinement

Instead of treating wet-lab validation as the final step, researchers can use experimental measurements to identify limitations in predicted properties and guide further sequence optimization.

For example, binding data can help refine models predicting antibody–antigen interactions, while expression and developability results can inform the selection of candidates with more balanced molecular properties.

The effectiveness of this approach depends on the availability of reliable experimental data and the speed at which it can be generated.

As computational antibody design continues to advance, integrated experimental platforms will play an important role in connecting in silico predictions with experimentally supported antibody candidates.

Biointron's RushData platform supports this process through rapid antibody production, high-throughput characterization, and structured data generation for iterative antibody discovery.

References:

1. Wu, F., et al. (2026). De novo design of epitope-specific antibodies via a structure-driven computational workflow. Nature Communications, 17, 625. https://doi.org/10.1038/s41467-025-67361-9