![[2602.18982] Conditionally Site-Independent Neural Evolution of Antibody Sequences](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.18982] Conditionally Site-Independent Neural Evolution of Antibody Sequences
Summary
The paper presents CoSiNE, a novel deep learning model for antibody sequence evolution that captures epistatic interactions and outperforms existing models in predicting variant effects.
Why It Matters
This research addresses a significant gap in antibody engineering by integrating evolutionary dynamics into deep learning models. By improving the prediction of antibody binding affinities, it has implications for vaccine development and therapeutic antibody design, which are critical in combating diseases.
Key Takeaways
- CoSiNE models antibody evolution using a continuous-time Markov chain.
- It captures complex epistatic interactions overlooked by traditional models.
- The model outperforms state-of-the-art language models in zero-shot variant effect prediction.
- Guided Gillespie sampling enhances optimization of antibody binding affinities.
- This approach could significantly advance therapeutic antibody development.
Computer Science > Machine Learning arXiv:2602.18982 (cs) [Submitted on 21 Feb 2026] Title:Conditionally Site-Independent Neural Evolution of Antibody Sequences Authors:Stephen Zhewen Lu, Aakarsh Vermani, Kohei Sanno, Jiarui Lu, Frederick A Matsen, Milind Jagota, Yun S. Song View a PDF of the paper titled Conditionally Site-Independent Neural Evolution of Antibody Sequences, by Stephen Zhewen Lu and 6 other authors View PDF HTML (experimental) Abstract: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 G...
