How Antibodies Learn: A Map of Beneficial Mutations ✎ express

Original: "Surveying the adaptive landscapes of 10,000 antibodies"

· Daniel PGH Wong, Aleksandra M. Walczak, Thierry Mora

arXiv:2606.21351 · 2026-06-19 · arXiv non-exclusive · independent analysis · 1 min

Scientists found a way to predict which mutations make antibodies stronger, simply by analyzing their evolution across different people.

The immune system is like a tireless editor, revising antibody drafts to make them stickier to viruses and bacteria. This process is called antibody maturation: random typos (mutations) arise in the genes, and natural selection keeps those that improve target capture.

The catch is that a mutation's "usefulness" isn't universal. What strengthens one antibody can ruin another. The authors analyzed over 10,000 "public clonotypes"—groups of very similar antibodies that arise independently in different people. By comparing mutations within these groups across 20 healthy donors, they spotted a pattern: frequent mutations usually give a small boost in strength, while rare ones give a large jump. It's a trade-off between reliability and ambition.

Common mutations are like minor edits an editor makes to almost every manuscript. Rare but powerful ones are like boldly rewriting a whole paragraph: it works only in special cases.

The new method requires no prior knowledge of antibody structure and works as a simple population-genetics tool. It was tested on antibodies against SARS-CoV-2 and flu — predictions matched real data. Moreover, popular protein language models mostly "see" neutral mutations, but after a simple renormalization, the same selection signals the method finds emerge.

The work showed that antibody evolution across different people follows similar paths, and these paths can be read as a map for developing more effective therapeutic antibodies.

Whole graph →

Antibodies can mutate at a rate a million times higher than ordinary genes: this allows them to quickly adapt to new threats.

In Peter Watts's novel "Blindsight," modified immune cells use directed evolution right inside a patient's body to defeat an unknown pathogen.

Concepts

predictive model genomics binding affinity genetic algorithm biological stochastic dynamics protein language model pattern recognition random forest medical image to genomic mapping immunity Hardy–Weinberg principle Hamilton's rule Large Language Models Baldwin effect Generative Adversarial Networks generalization

I am the author

How Antibodies Learn: A Map of Beneficial Mutations ✎ express