How Evolution Teaches Antibodies express
An antibody is a protein that catches viruses and bacteria. When an enemy enters the body, immune cells start trying out different antibody variants, as if testing different keys against a lock. Beneficial changes are kept, harmful ones are discarded. Over a few weeks, this process yields an antibody that fits its target perfectly.
Conventional computer models don't see this history. They look at finished antibodies and treat them as independent, like cards in a deck. But each antibody is a descendant of a previous one, and its structure records a path of trial and error. The new model CoSiNE reads this lineage. It represents antibody evolution as a chain of small steps, where each step is a single-letter substitution in the genetic code.
The model doesn't just replay the past: it can accelerate it. An embedded "guide" mechanism steers the search toward antibodies that bind more tightly to the desired target. It's like a plant breeder who doesn't wait for random mutations but directly plants seeds with desired traits. That's why CoSiNE outperforms even the most advanced language models, which predict protein properties but don't understand where those properties come from.
Original abstract
as written by the authors arXiv:2602.18982
Common deep learning approaches for antibody engineering focus on modeling the marginal distribution of sequences. By treating sequences as independent samples, however, these methods overlook affinity maturation as a rich and largely untapped source of information about the evolutionary process by which antibodies explore the underlying fitness landscape. In contrast, classical phylogenetic models explicitly represent evolutionary dynamics but lack the expressivity to capture complex epistatic interactions. We bridge this gap with CoSiNE, a continuous-time Markov chain parameterized by a deep neural network. Mathematically, we prove that CoSiNE provides a first-order approximation to the intractable sequential point mutation process, capturing epistatic effects with an error bound that is quadratic in branch length. Empirically, CoSiNE outperforms state-of-the-art language models in zero-shot variant effect prediction by explicitly disentangling selection from context-dependent somatic hypermutation. Finally, we introduce Guided Gillespie, a classifier-guided sampling scheme that steers CoSiNE at inference time, enabling efficient optimization of antibody binding affinity toward specific antigens.
A single antibody can change its shape to grab two different targets, like a hand that adapts to different objects.