The Inaugural International AIntibody Challenge: Putting AI-Designed Antibodies to the Test

Artificial intelligence (AI) is rapidly transforming drug discovery, but how well can AI actually design functional antibodies? The inaugural International AIntibody Challenge set out to answer that question through a first-of-its-kind global competition focused on evaluating AI and machine learning platforms for antibody design.

The challenge attracted an impressive 511 antibody submissions from 29 organizations, representing a diverse mix of academic institutions, non-profit research organizations, biotechnology companies, AI-driven drug discovery firms, and global pharmaceutical companies. This broad participation highlighted the growing interest in leveraging AI/ML to accelerate antibody discovery and optimization.

The Three Challenges

Participants were tasked with solving three key antibody engineering challenges that reflect real-world drug discovery hurdles.

Challenge 1: In Silico Antibody Affinity Maturation

In the first challenge, participants were asked to design antibody complementarity-determining regions (CDRs) with improved affinity for a specified target from NGS datasets. The goal was to modify the sequence to affinity mature the antibodies and improve binding.

Challenge 2: In Silico Affinity Rank Prediction for Antibody Discovery

The second challenge tested predictive rather than generative skill. Using NGS data, participants had to identify the highest-affinity antibody sequences within three heavy-chain CDR3 (HCDR3) clusters, 27F, 28F, and 47F, without the benefit of wet-lab feedback.

Challenge 3: NGS-Inspired Computational Antibody Design

The third challenge employed a fully computational de novo design strategy requiring antibody discovery without prior library-based guidance. Participants generated antibody sequences with no prior exposure, modifying only the CDR regions while keeping framework sequences fixed, with success measured by both binding affinity and developability properties.

Turning AI Predictions into Real-World Results

A key differentiator of the AIntibody Challenge was its commitment to experimental validation. Computational predictions alone are not enough; designed antibodies must perform in the laboratory.

To validate contestants’ predictions, Specifica partnered with GENEWIZ to leverage its high-throughput antibody production capabilities to synthesize, express, and purify hundreds of AI-designed antibody candidates for downstream testing by Carterra, Mosaic, and Sapidyne.

Antibody Production, Testing, and Analysis Workflow: A Five Step Process

Step 1: Antibody Production and Developability Assessment

GENEWIZ produced the submitted antibody candidates and shipped them to Carterra for Surface Plasmon Resonance (SPR) binding analysis and Mosaic Biosciences developability assessment

This phase established both binding performance and manufacturability characteristics for each antibody candidate.

Step 2: SPR-Based Candidate Ranking

Antibodies were ranked according to their SPR performance.

Candidates exhibiting poor or borderline developability characteristics were prioritized for additional evaluation using KineExA technology at Sapidyne to obtain deeper insights into binding behavior.

Step 3: KineExA Single-Point Screening

Sapidyne Instruments conducted single-point KineExA screening on selected candidates to confirm that rankings generated by KineExA were consistent with the initial SPR-derived rankings.

Step 4: Comprehensive KineExA Analysis

Up to five of the highest-priority candidates with developability scores of ≤3 advanced to full KineExA characterization.

This stage provided highly sensitive affinity measurements and enabled further differentiation among top-performing antibodies.

Step 5: Final Candidate Ranking

Data from the SPR analysis, Developability assessment, and KineExA screening were integrated to generate final rankings and identify the strongest overall antibody candidates.

The AIntibody Challenge demonstrated the power of collaboration across the biotechnology ecosystem. Multiple organizations contributed specialized expertise throughout the workflow:

• Specifica administered and coordinated the challenge.
• GENEWIZ produced AI-designed antibody candidates for experimental validation.
• Carterra performed high-throughput SPR analysis.
• Mosaic Biosciences assessed antibody developability characteristics.
• Sapidyne Instruments conducted KineExA measurements to evaluate ultra-high-affinity interactions.
• Bio-Techne and the La Jolla Institute for Immunology provided critical antigen support.

Together, these organizations created one of the most comprehensive experimental evaluations of AI-generated antibodies conducted to date.

Key Findings

Challenge 1: Affinity Maturation

The challenge showcased notable successes, with multiple teams designing developable antibodies exhibiting sub-100 pM affinities; however, the affinity improvements were not consistent across all challenges. These findings provide strong evidence that AI can deliver meaningful improvements in antibody design, while also revealing opportunities to strengthen performance consistency across a wider range of discovery challenges.

Challenge 2: Affinity Ranking Prediction

While computational approaches showed promise in modeling affinity-matured antibodies, accurately pinpointing the top-performing clones within HCDR3 sequence clusters proved considerably more difficult. Only one approach consistently exceeded the performance expected from random clone selection.

Challenge 3: Novel Antibody Design

Results from the novel antibody design challenge highlighted both the potential and current limitations of AI-driven discovery. Although several approaches produced encouraging candidates, consistently designing entirely new antibodies with optimal affinity and developability remains an important area for continued innovation.

What Did We Learn?

The AIntibody Challenge delivered one of the most rigorous real-world evaluations of AI-driven antibody discovery to date, highlighting where current technologies excel and where significant challenges remain.

The results demonstrated that AI/ML can successfully optimize antibodies within well-defined and biologically grounded design spaces. At the same time, the challenge revealed important limitations that remain to be addressed, particularly in affinity prediction and novel antibody generation with favorable developability.

Ultimately, these findings offer valuable guidance for the next generation of AI/ML models and workflows aimed at accelerating biologics discovery.

Looking Ahead

Building on the success of this inaugural event, the Antibody Society will administer future AIntibody Challenges, expanding opportunities for the scientific community to benchmark, improve, and validate AI technologies for antibody discovery.

As AI continues to evolve, initiatives like the AIntibody Challenge will play a critical role in determining where machine learning delivers meaningful advances and where additional innovation is still needed.

Learn more about the AIntibody Challenge

Erasmus, M.F., Bedinger, D., Hopkins, E. et al. A blinded, prospective benchmark of in silico antibody discovery anchored to experimental affinity and developability. Nat Biotechnol (2026). https://doi.org/10.1038/s41587-026-03238-6

About the Author

Crystal holds a Ph.D. from the University of Virginia and brings over a decade of experience in clinical immunology, infectious disease research, and business development. She has led antibody manufacturing initiatives and supported high‑throughput clinical vaccine testing.

At GENEWIZ, as the subject matter expert for recombinant antibody production and viral packaging services, Crystal helps shape strategic partnerships, drives business development, and supports the growth of Global Synthesis Solutions, always delivering a customer‑focused experience grounded in scientific rigor, quality, and care.