Antibody Discovery
and Engineering

Antibody discovery has historically relied on two main approaches: immunization and in vitro display. In immunization-based discovery, an animal (typically mouse, rat, rabbit, or camelid) is immunized with the target antigen, and the immune system generates a diverse repertoire of antibodies through natural affinity maturation. B cells producing high-affinity antibodies are then isolated by hybridoma technology, single B-cell sorting, or repertoire sequencing. This approach leverages the immune system's optimization machinery and typically yields antibodies with nanomolar to sub-nanomolar affinities, though with limited control over epitope selection.

In vitro display technologies—phage display, yeast display, ribosome display, and mammalian display—bypass the immune system entirely. A synthetic or semi-synthetic library of antibody variants (typically 10⁹ to 10¹² unique sequences) is displayed on the surface of phage particles, yeast cells, or ribosomes, and binders are enriched through iterative rounds of selection (panning) against the target. Display approaches offer precise control over selection conditions, enable counter-selection against undesired targets, and can discover antibodies against targets that are poorly immunogenic or toxic to animals. However, initial affinities from naïve libraries are often in the micromolar range and require subsequent affinity maturation.

A third approach—computational de novo antibody design—has emerged as a viable discovery strategy. Tools like RFDiffusion, ProteinMPNN, and deep learning-based structure predictors (AlphaFold2, Boltz-2, ESMFold) enable the generation of antibody sequences and structures entirely in silico, without immunization or library screening. The process typically involves generating backbone structures that complement the target epitope, designing sequences that fold into those backbones, and scoring candidates for predicted binding affinity, structural confidence, and developability.

Computational approaches are particularly powerful for nanobodies (VHH single-domain antibodies), which have a simpler architecture than conventional IgG antibodies and are more amenable to computational modeling. VHH domains are approximately 15 kDa, have a single variable domain, and derive their binding specificity primarily from three CDR loops—especially the elongated CDR3 that is characteristic of camelid heavy-chain-only antibodies. Their small size, high stability, and ease of production make them attractive both as standalone therapeutics and as building blocks for multi-specific constructs.

A good antibody candidate is not just one that binds the target tightly. Affinity matters, but so do specificity (absence of off-target binding), developability (expression level, aggregation propensity, viscosity, thermal stability), and manufacturability (sequence liabilities such as deamidation sites, unpaired cysteines, and glycosylation motifs). The best discovery campaigns evaluate all of these properties early, not just affinity, so that lead candidates entering optimization have a realistic path to the clinic. Increasingly, computational pre-screening for developability is being integrated directly into the discovery workflow—filtering out problematic sequences before any protein is expressed, saving significant time and cost.