How structural biology, antibody engineering and AI are advancing modern antibody drug discovery
By Dr. Neil Taylor
Monoclonal antibodies are now among the best-selling medicines in the world. They treat cancers, autoimmune diseases, migraine, high cholesterol and infections. More than 100 antibody therapeutics have been approved by the US FDA, and hundreds more are in clinical development.
Antibody drug discovery has changed considerably since the first monoclonal antibodies were made in the 1970s. This article looks at the modern antibody discovery workflow and the innovative techniques changing each stage of the process.
Why antibodies work so well
An antibody is a Y-shaped protein. The tips of the Y carry six short loops called complementarity-determining regions (CDRs). Together, these loops form a binding surface that can recognise a specific target, known as an antigen, with very high precision.
This precision is the main appeal. Antibodies can bind targets that small-molecule drugs struggle with, such as large, flat protein surfaces. Their specificity also tends to mean fewer off-target effects.
The antibody discovery workflow
Most antibody discovery programmes follow a broadly similar sequence:
- Target and antigen preparation: Producing the target protein, or the relevant part of it, in a form that can be used for screening.
- Generating diversity: Creating a large pool of candidate antibodies.
- Screening and selection: Finding the candidates that bind the target.
- Characterisation: Measuring affinity, specificity and function, and identifying where on the target each antibody binds (the epitope).
- Optimisation: Improving binding, stability and manufacturability, and reducing the risk of an immune response.
- Developability assessment: Checking that the antibody can be produced, formulated and stored at scale.
Generating and screening antibody candidates
There are two broad routes to generating diversity in modern antibody discovery.
Immunisation-based approaches use the immune system itself. Animals, often transgenic mice engineered to produce human antibodies, are immunised with the antigen. Their B cells are then harvested. Single B-cell screening technologies, including microfluidic platforms, now let researchers isolate and sequence individual antibody-producing cells at high throughput.
Display-based approaches use antibody library discovery platforms. Phage display, yeast display and mammalian display present billions of different antibody fragments on the surface of a virus or cell. Rounds of selection enrich for those that bind the target. Synthetic libraries can be designed to favour properties such as stability from the outset.
Next-generation sequencing has transformed both routes. Rather than examining a few hundred clones, teams can now read millions of antibody sequences and track which ones are enriched during selection.

Structural biology: seeing how antibodies bind
Knowing exactly where and how an antibody binds is increasingly central to antibody discovery and engineering.
Structural information helps teams:
- Choose antibodies that block the right part of the target.
- Avoid epitopes shared with related proteins, reducing off-target risk.
- Design improvements to the complementarity-determining regions (CDRs) with confidence.
- Build a stronger intellectual property position around a defined epitope.
X-ray crystallography and cryo-electron microscopy (cryo-EM) remain the definitive methods for determining antibody–antigen complex structures. Hydrogen–deuterium exchange mass spectrometry offers a faster, lower-resolution view of the epitope.
The role of AI and computational antibody design
Computational methods are now part of most modern antibody discovery platforms. Several techniques are worth noting:
- Structure prediction: Specialised AI models can predict antibody structures from sequence, with the framework regions predicted well. The long, variable CDR-H3 loop, often the most important for binding, remains the hardest part to model accurately.
- Complex prediction: Newer co-folding models can propose how an antibody and antigen fit together. Accuracy is improving but is still less reliable than for many other protein complexes.
- Developability prediction: Models trained on historical data can flag sequences likely to aggregate, be poorly expressed or behave badly in formulation, before money is spent making them.
- Generative design: AI methods are beginning to propose new antibody sequences or binders designed for a chosen epitope. Early results are promising, and experimental validation remains essential.
Data is the common thread
Every one of these antibody discovery techniques generates data: sequences, binding measurements, structures and developability scores.
The organisations moving fastest connect these data types so that each new programme learns from the last. A structure of an antibody–antigen complex is far more valuable when it can be searched alongside every related complex the organisation has solved.
About DesertSci
DesertSci's Proasis platform manages and analyses protein structures at scale, including protein–protein and antibody–antigen complexes, for many of the world's leading pharmaceutical companies.
Learn more at desertsci.com and follow Dr Neil Taylor on LinkedIn for more insights.



