Accurate de novo peptide sequencing using fully convolutional neural networks

Area of Science:

• Proteomics and Bioinformatics
• Computational Biology
• Mass Spectrometry Analysis

Background:

• De novo peptide sequencing is crucial for identifying novel peptides without relying on databases.
• Existing de novo sequencing algorithms face challenges with accuracy and coverage, limiting their use in proteomics.
• There is a need for more accurate and efficient de novo sequencing methods.

Purpose of the Study:

• To introduce PepNet, a fully convolutional neural network designed for high-accuracy de novo peptide sequencing.
• To evaluate PepNet's performance against current state-of-the-art algorithms.
• To demonstrate PepNet's utility as a complementary tool for large-scale proteomics data analysis.

Main Methods:

• PepNet utilizes a fully convolutional neural network architecture.
• The model takes MS/MS spectra as input and outputs optimal peptide sequences with confidence scores.
• Training involved 3 million high-energy collisional dissociation MS/MS spectra from human peptide libraries.

Main Results:

• PepNet significantly outperforms existing algorithms like PointNovo and DeepNovo in both peptide-level and positional-level accuracy.
• PepNet successfully sequences a substantial number of spectra missed by traditional database search engines.
• PepNet demonstrates superior computational efficiency, running 3x and 7x faster than PointNovo and DeepNovo on GPUs, respectively.

Conclusions:

• PepNet represents a significant advancement in de novo peptide sequencing accuracy and efficiency.
• The algorithm can effectively identify novel peptides and complement existing database search methods in proteomics.
• PepNet's speed and accuracy make it suitable for analyzing large-scale proteomics datasets.