Protein identification with deep learning: from abc to xyz (1710.02765v1)

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).

• The paper introduces DeepNovo, a hybrid framework that combines de novo peptide sequencing and database search to achieve an 89.8% peptide recall rate.
• The paper employs CNN and LSTM networks to model amino acid sequences and fragment ion intensities, significantly enhancing prediction accuracy.
• The paper’s data-centric approach leverages extensive protein databases and beam search algorithms, setting a new benchmark in proteomic analysis.

Summary of "Protein identification with deep learning: from abc to xyz"

This paper introduces DeepNovo, a sophisticated deep learning tool designed for protein identification from tandem mass spectrometry data. The haLLMark of DeepNovo is its integration of both de novo peptide sequencing and database search strategies into a unified framework, leveraging advancements in deep learning, particularly convolutional neural networks (CNN) and long short-term memory (LSTM) networks. The framework capitalizes on the growing availability of extensive protein databases to enhance the accuracy and efficiency of peptide identification.

Core Contributions

DeepNovo adopts a data-centric approach, shifting away from traditional algorithm-centric methods by utilizing large datasets to fuel its learning capabilities. It applies a hybrid solution to peptide identification by simultaneously utilizing a scoring function common to both de novo sequencing and database search. The research specifically evaluates the performance of the DeepNovo tool on a Saccharomyces cerevisiae proteome dataset, acquired from a high-resolution Orbitrap Fusion mass spectrometer.

Methodological Advancements

1. DeepNovo Scoring Function: The intricate scoring function is based on calculating conditional probabilities of amino acid sequences given spectral data. It employs CNNs to discern intensity distributions of fragment ions and LSTMs to understand peptide sequence patterns. This is critical for enhancing the prediction accuracy of amino acid sequences.
2. Database Search: DeepNovo optimally narrows down candidate sequences from the UniProt database by performing in silico digestion and reducing the search space based on peptide mass and cleavage rules. The use of bi-directional sequencing augments its ability to handle missing fragment ions.
3. De Novo Sequencing: In the challenge of de novo sequencing, where no database reference is available, the paper demonstrates the application of a beam search algorithm for iterative sequence assembly, incorporating dynamic programming methods to address mass-match constraints.
4. Hybrid Approach: By integrating database search and de novo sequencing, DeepNovo achieves a higher degree of versatility. The approach predicts novel peptide sequences that potentially outperform existing database entries, thus offering a comprehensive solution to proteomic analysis.

Evaluation and Results

The performance assessment reveals that DeepNovo achieves a substantial improvement in peptide recall rates compared to existing methods. Specifically, a peptide recall rate of 89.8% was reported, indicating high identification accuracy. The robustness of the hybrid system is further emphasized by its ability to predict novel peptide sequences that diverge from those identified by traditional methods, such as PEAKS, with examples showcasing its enhanced predictive capability.

Implications and Future Directions

The integration of deep learning in peptide identification marks a significant advancement in the field of proteomics. It promises not only to improve identification accuracy but also to accelerate the processing of vast proteomics data sets. Practically, this methodology could be pivotal in advancing biomedical research, opening pathways to more thorough analyses of proteome transformations under various biological conditions.

Looking forward, further enhancement of the DeepNovo framework could involve integrating more sophisticated feature extraction models or leveraging more expansive and diverse training datasets. Additionally, addressing the computational challenges associated with scaling such techniques for larger datasets and more complex biological samples is a promising direction for advancing this research domain. The adaptability of this framework also presents opportunities for cross-disciplinary integration, expanding its utility in various fields of bioinformatics and systems biology.