November 2025
How ML & AI Are Changing Antibody Discovery
AI and machine learning are transforming how we discover and design antibodies, from computational modelling and generative design to structure prediction and developability optimisation. At Antibody Engineering & Therapeutics US 2025, experts will reveal how these technologies are reshaping antibody R&D and accelerating the next wave of therapeutic innovation.
1. Computational Design Is Accelerating Discovery
For decades, antibody discovery relied heavily on wet-lab screening, vast libraries, months of benchwork, and low hit-to-lead efficiency. Today, computational design tools can model binding interfaces, simulate antigen-antibody interactions, and prioritise candidates long before experimental validation.
AI-driven computational workflows are enabling:
- In silico screening to narrow libraries to the most promising binders
- Affinity and specificity optimisation through sequence-level prediction
- Developability profiling, predicting aggregation, stability, and manufacturability
These tools reduce both cost and time-to-lead dramatically.
At the conference, this evolution is exemplified in sessions like “An Integrated Pipeline from Antibody Panel to Drug Candidates” which demonstrates how AI is paired with high-throughput expression and developability assays, and “Advancing Multi-Objective Antibody Selection Through AI-Driven Pareto Optimisation”, showcasing how teams manage complex property trade-offs computationally.
Computational design is no longer a supporting tool, it’s the foundation of modern discovery pipelines, and these sessions show how AI/ML now sit at the core of end-to-end antibody workflows.
2. Generative AI Is Designing Antibodies De Novo
Generative models, the same architectures behind modern large language models, are now creating antibodies from scratch. Trained on massive datasets of known antibodies and antigen structures, these models can propose novel variable regions, CDR loops, or entire frameworks optimised for desired binding properties.
At the frontier are:
- Diffusion and transformer models generating sequence variants that “look” and “act” like high-affinity antibodies
- Reinforcement learning frameworks optimising binding, stability, and immunogenicity simultaneously
- Foundation models for proteins that can generalise across antibody classes
This is enabling researchers to explore entirely new parts of sequence-space, accelerating the move from “discovery” to “design.”
The session “Tackling Challenging Targets with an End-to-End AI-driven Antibody Discovery Platform” will present case studies where AI platforms have generated leads for difficult targets, a real-world demonstration of generative design in practice.
Generative AI has moved from theoretical promise to operational reality. This session shows how AI-driven design is already producing viable therapeutic leads.
3. Structure Prediction Is Closing the Loop Between Sequence and Function
The arrival of deep-learning structure predictors (e.g., AlphaFold, ESMFold) has revolutionised protein design. For antibodies, accurate structural modelling of variable regions, especially the flexible CDR-H3 loop, was historically a bottleneck.
Now, next-generation ML tools can:
- Predict 3D structures with near-experimental accuracy
- Model antibody–antigen complexes
- Enable rational design of paratopes for multispecifics and conditionally active formats
These advances feed directly into protein engineering workflows, guiding which designs to test, how to improve manufacturability, and how to combine binding domains into bispecific or multispecific antibodies.
A session under Engineering Innovative Formats & Scaffolds, “Selective Multifunctional Antibodies to Combat Disease” illustrates how modern structure prediction underpins the design of conditionally active and multifunctional antibodies.
Structure prediction is now an enabler that links sequence to structure to function. These advances make innovative formats and multispecific designs practical at scale.
4. AI + Experimental Synergy: The Hybrid Discovery Workflow
AI is not replacing the lab; it’s re-wiring it. The most successful discovery pipelines are hybrid systems, where algorithms generate hypotheses and experiments validate them rapidly.
This synergy enables:
- Rapid iteration between design, expression and testing
- Integration of real-world experimental data to continually refine AI models
- Closed-loop discovery, “design, build, test, learn” in days rather than months
The Pre-Conference Workshop on “Predicting Antibody Developability Using Interpretable Machine Learning” highlights how teams can integrate ML-based developability predictions early in the workflow, ensuring that AI and wet-lab steps remain tightly connected.
The agenda reflects a shift toward unified AI–experimental systems, discovery is now a collaborative loop between models and the bench, not a linear process.
5. From Discovery to Developability: AI Across the Pipeline
AI now spans the entire antibody lifecycle:
- Discovery: identifying hits and generating binders
- Engineering: predicting structure, binding and biophysical behaviour
- Optimization: tuning affinity, stability and solubility
- Manufacturing: forecasting aggregation and expression issues
- Clinical translation: modelling pharmacokinetics and immunogenicity
This holistic integration marks a new era of data-driven biologics. It’s not just faster discovery, it’s smarter discovery.
The keynote “How Multispecific Biologics Are Transforming Pharmacotherapy” emphasises that next-generation modalities require coordinated advances across AI, engineering, and manufacturability, reinforcing the need for integrated pipelines.
AI is now embedded across the full antibody development lifecycle. The conference makes it clear: success in 2025 depends on how effectively teams integrate data, models, and molecular intelligence, not just on libraries or assays.
Why It Matters for Antibody Engineers and Therapeutic Developers
As modalities evolve, from bispecifics to conditionally active formats and novel scaffolds, AI and ML will be the engines that make those designs feasible.
Sessions at the Antibody Engineering & Therapeutics US 2025 conference will showcase exactly how these computational approaches are being applied, refined, and scaled across the industry.
In 2025, success in antibody R&D won’t depend solely on access to libraries or assays, but on how effectively teams harness data, models, and molecular intelligence