[Paper Review] Protein identification with deep learning: from abc to xyz

Ngoc Hieu Tran, Zachariah Levine+3 more|arXiv (Cornell University)|Oct 8, 2017

Machine Learning in Bioinformatics17 references3 citations

TL;DR

This paper introduces DeepNovo, a deep learning framework that integrates de novo peptide sequencing and database search using convolutional and recurrent neural networks to identify proteins from tandem mass spectrometry data. By combining sequence prediction with de Bruijn graph assembly, DeepNovo achieves high-accuracy protein identification, with open-source models and tutorials available on GitHub for reproducibility and broader adoption in proteomics research.

ABSTRACT

Proteins are the main workhorses of biological functions in a cell, a tissue, or an organism. Identification and quantification of proteins in a given sample, e.g. a cell type under normal/disease conditions, are fundamental tasks for the understanding of human health and disease. In this paper, we present DeepNovo, a deep learning-based tool to address the problem of protein identification from tandem mass spectrometry data. The idea was first proposed in the context of de novo peptide sequencing [1] in which convolutional neural networks and recurrent neural networks were applied to predict the amino acid sequence of a peptide from its spectrum, a similar task to generating a caption from an image. We further develop DeepNovo to perform sequence database search, the main technique for peptide identification that greatly benefits from numerous existing protein databases. We combine two modules de novo sequencing and database search into a single deep learning framework for peptide identification, and integrate de Bruijn graph assembly technique to offer a complete solution to reconstruct protein sequences from tandem mass spectrometry data. This paper describes a comprehensive protocol of DeepNovo for protein identification, including training neural network models, dynamic programming search, database querying, estimation of false discovery rate, and de Bruijn graph assembly. Training and testing data, model implementations, and comprehensive tutorials in form of IPython notebooks are available in our GitHub repository (https://github.com/nh2tran/DeepNovo).

Motivation & Objective

• To develop a unified deep learning framework that combines de novo sequencing and database search for improved protein identification from tandem mass spectrometry data.
• To enhance accuracy and efficiency in peptide identification by leveraging neural networks trained on spectral data.
• To integrate de Bruijn graph assembly into the deep learning pipeline for reconstructing full protein sequences from fragmented peptide data.
• To provide a complete, reproducible workflow including model training, false discovery rate estimation, and dynamic programming search.
• To make training data, model implementations, and tutorials publicly available via a GitHub repository to support adoption in proteomics research.

Proposed method

• Employing convolutional neural networks (CNNs) to extract features from tandem mass spectrometry (MS/MS) spectra.
• Using recurrent neural networks (RNNs), particularly bi-LSTM layers, to model sequential dependencies in amino acid prediction.
• Designing a joint architecture that combines de novo sequencing and database search into a single end-to-end framework.
• Applying dynamic programming to refine peptide sequence candidates based on spectral alignment scores.
• Integrating de Bruijn graph assembly to reconstruct full protein sequences from identified peptides.
• Estimating false discovery rates using target-decoy database strategies within the deep learning pipeline.

Experimental results

Research questions

• RQ1 Can a unified deep learning model effectively combine de novo sequencing and database search to improve peptide identification accuracy?
• RQ2 How does the integration of de Bruijn graph assembly enhance protein sequence reconstruction from MS/MS data?
• RQ3 To what extent does the proposed end-to-end deep learning framework outperform traditional methods in protein identification tasks?
• RQ4 Can the framework achieve high performance while maintaining interpretability and reproducibility through open-source tools and tutorials?

Key findings

• DeepNovo achieves state-of-the-art performance in de novo peptide sequencing by leveraging CNNs and RNNs to predict amino acid sequences directly from MS/MS spectra.
• The integration of database search within the deep learning framework significantly improves identification accuracy compared to standalone de novo methods.
• The use of de Bruijn graph assembly enables robust reconstruction of full protein sequences from peptide identifications, enhancing biological interpretability.
• The framework supports reliable false discovery rate estimation, a critical requirement for statistical validation in proteomics.
• Comprehensive training data, model weights, and IPython notebook tutorials are publicly available, enabling reproducibility and community adoption.