AI Is Now Designing Antibodies From Scratch

The Old Way of Making Biologics Was Painfully Slow

Biologic drugs — antibodies, proteins, peptides — are among the most effective therapies in modern medicine. They treat cancers, autoimmune diseases, and rare genetic conditions. But developing a single biologic has traditionally taken 10 to 15 years and cost upwards of $2 billion.

The bottleneck was always the same: trial and error. Researchers would screen thousands of candidate molecules, test them in the lab, optimize their binding properties, and repeat. Each cycle took months. Most candidates failed.

That process is now being replaced by something fundamentally different. AI systems can design functional antibodies and proteins computationally, before a single molecule is ever synthesized in a lab.

How AI Designs Proteins Without a Lab

The breakthrough started with protein structure prediction. DeepMind’s AlphaFold proved that neural networks could accurately predict how amino acid sequences fold into three-dimensional structures. That was the foundation. What followed has been far more ambitious.

Large language models trained on protein sequences — sometimes called protein LLMs — now generate entirely new molecules. These models learn the grammar of biology: which sequences fold stably, which bind tightly to specific targets, which avoid triggering immune responses.

The result is that AI can now design novel antibodies, peptides, and engineered proteins with specific functions. It can optimize binding affinity, thermal stability, and manufacturability simultaneously — tasks that used to require separate teams working for years.

David Baker’s lab at the University of Washington, which won the 2024 Nobel Prize in Chemistry for computational protein design, has continued pushing boundaries. Their tools can now generate proteins that don’t exist in nature but perform precisely defined biological functions.

Real Money Is Flowing In

The investment numbers tell the story. In March 2026, Earendil Labs raised $787 million specifically to build an AI platform for antibody and biologic design. That’s not a seed round — it’s a signal that institutional investors see this as a proven approach, not a speculative bet.

Eli Lilly partnered with NVIDIA to build a dedicated AI supercomputer for drug discovery and manufacturing. The company also invested $250 million in an AI research partnership with Purdue University. Bayer selected Cradle’s AI software platform for protein engineering, aiming to reduce the optimization cycles needed to bring molecules into clinical development.

Roche deployed more than 3,500 NVIDIA Blackwell GPUs across its global operations, powering biological foundation models that accelerate everything from target identification to molecular optimization.

These aren’t pilot programs. They’re core infrastructure investments by companies that generate tens of billions in annual biologic drug revenue.

AI-Designed Biologics Are Entering the Clinic

The ultimate test for any drug design approach is whether it produces molecules that work in humans. On that front, 2026 marks a turning point.

Early AI-designed biologics — including peptide therapeutics, engineered antibodies, and mRNA-based candidates — are now entering clinical evaluation. These aren’t molecules that were merely optimized by AI. They were conceived computationally, with AI making the initial design decisions about sequence, structure, and function.

The shift matters because it compresses timelines dramatically. Traditional antibody discovery campaigns that took 18 to 24 months can now be completed in weeks. Companies report reducing the number of experimental optimization cycles from dozens to fewer than five.

STAT News reported in January 2026 that AI has “finally started making drug-like antibodies,” noting that the quality gap between computationally designed and traditionally discovered molecules has closed significantly.

Multispecific Antibodies: Where AI Really Shines

One area where AI provides an outsized advantage is multispecific antibody design. These are antibodies engineered to bind two or more targets simultaneously — a capability that’s extremely valuable for treating complex diseases like cancer.

Designing multispecific antibodies by conventional methods is enormously difficult. Each additional binding site multiplies the possible configurations. The design space becomes too large for experimental screening alone.

AI handles this combinatorial complexity naturally. Models can evaluate billions of possible configurations and predict which combinations will fold correctly, bind their targets, and remain stable in manufacturing conditions. What used to be one of the hardest problems in biologics engineering is becoming routine.

The 2026 biologics pipeline increasingly features bispecific and trispecific antibodies that would have been impractical to develop without computational design.

What AI Still Can’t Do Well

For all the progress, significant gaps remain. Current AI models excel at predicting molecular structure but struggle with the full complexity of biological systems.

Immunogenicity — whether a biologic will trigger an unwanted immune response in patients — remains difficult to predict computationally. So do pharmacokinetics: how the drug distributes through the body, how quickly it’s cleared, and whether it reaches its target tissue at therapeutic concentrations.

Cellular context is another blind spot. A molecule that binds perfectly to a purified target protein in simulation may behave differently inside a living cell, where crowding, pH variations, and competing interactions all affect performance.

These limitations mean that AI-designed biologics still require extensive experimental validation. The technology accelerates the design phase but doesn’t eliminate the need for lab work and clinical trials.

The Novo Nordisk-OpenAI Partnership

Perhaps the most telling development of 2026 is Novo Nordisk’s partnership with OpenAI, announced in April. The Danish pharmaceutical giant — best known for its GLP-1 drugs like Ozempic — is working with OpenAI to analyze complex biological datasets, identify promising drug candidates, and reduce development timelines.

This partnership is significant because Novo Nordisk is a biologics-focused company. Its core products are peptide hormones and protein therapeutics. Bringing OpenAI’s general-purpose AI capabilities to bear on biologics-specific problems suggests that the field is moving beyond specialized protein design tools toward broader AI integration.

The collaboration also reflects a growing recognition that the bottleneck in biologics development isn’t just molecular design — it’s the ability to process and interpret the massive datasets generated by genomic, proteomic, and clinical research.

What This Means for Patients

The practical impact is straightforward: more biologic drugs, developed faster, targeting diseases that currently lack effective treatments.

AI-designed antibodies could expand treatment options for cancers that don’t respond well to existing therapies. Computationally optimized proteins could enable gene therapies with better safety profiles. Faster development cycles mean that when new disease targets are identified, drugs can reach patients in years rather than decades.

The cost implications are significant too. If AI can reduce the failure rate in biologics development — currently around 90% of candidates fail in clinical trials — the savings would cascade through the entire drug pricing chain.

None of this happens overnight. But the infrastructure is being built, the capital is flowing, and the first AI-designed biologics are in human trials. The question isn’t whether AI will reshape biologics development. It’s how quickly the industry can adapt to a world where the best drug designers are algorithms.

Related reading: How AI Cut Drug Discovery Costs From Billions to Millions | AI-Designed Drugs Hit Phase III Trials