Scholarlink

We present a novel method for predicting multiple reaction monitoring (MRM) transitions for peptides in targeted proteomics. Our approach employs a hash-based representation inspired by convolutional neural networks, efficiently encoding peptide fragments as sparse count vectors that capture local sequence context. Using gradient-boosted decision trees, our method achieves mean Hits@5 scores of 3.4318 (hash-based) and 3.5405 (hybrid model with target frequency), significantly outperforming baselines. Transpiling trained models into Zig enables exceptional computational efficiency, with low memory usage (1180 kB) and a throughput of 388–451 peptides/second even on mobile devices, enabling lightweight, high-speed processing for scalable peptide MRM transition prediction in high-throughput proteomics workflows.

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

Efficient Peptide MRM Transition Prediction via Convolutional Hashing

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ramon Adàlia,
  • Gemma Sanjuan,
  • Tomàs Margalef,
  • Ismael Zamora

摘要

We present a novel method for predicting multiple reaction monitoring (MRM) transitions for peptides in targeted proteomics. Our approach employs a hash-based representation inspired by convolutional neural networks, efficiently encoding peptide fragments as sparse count vectors that capture local sequence context. Using gradient-boosted decision trees, our method achieves mean Hits@5 scores of 3.4318 (hash-based) and 3.5405 (hybrid model with target frequency), significantly outperforming baselines. Transpiling trained models into Zig enables exceptional computational efficiency, with low memory usage (1180 kB) and a throughput of 388–451 peptides/second even on mobile devices, enabling lightweight, high-speed processing for scalable peptide MRM transition prediction in high-throughput proteomics workflows.