Scholarlink

To create effective vaccines and diagnostic tools, as well as to study how viruses evolve and adapt to host immune responses, viral antigens must be identified. Here we introduce a novel probabilistic model (Vegvisir) for the prediction of viral T-cell epitopes based on the architecture of a Variational Autoencoder. In this work, we illustrate how the model learns meaningful sequence latent representations from which viral peptides can be classified according to their immunogenicity. We also explore how these representations provide insights into the biological properties of epitopes. Likewise, we show how this model outperforms the current state-of-the-art methods on a well-curated dataset. Finally, we perform different experiments to carefully stress-test and assess Vegvisir for biases to monitor its capability to generalize.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Vegvisir: Probabilistic Model (VAE) for Viral T-Cell Epitope Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Lys Sanz Moreta,
  • Ibel Carri,
  • Heli M. Garcia Alvarez,
  • Morten Nielsen

摘要

To create effective vaccines and diagnostic tools, as well as to study how viruses evolve and adapt to host immune responses, viral antigens must be identified. Here we introduce a novel probabilistic model (Vegvisir) for the prediction of viral T-cell epitopes based on the architecture of a Variational Autoencoder. In this work, we illustrate how the model learns meaningful sequence latent representations from which viral peptides can be classified according to their immunogenicity. We also explore how these representations provide insights into the biological properties of epitopes. Likewise, we show how this model outperforms the current state-of-the-art methods on a well-curated dataset. Finally, we perform different experiments to carefully stress-test and assess Vegvisir for biases to monitor its capability to generalize.