Deep Convolutional Neural Networks Help Scoring Tandem Mass Spectrometry Data in Database-Searching Approaches.

Area of Science:

• Proteomics
• Computational Biology
• Spectrometry

Background:

• Spectrum annotation in mass spectrometry (MS)/MS is difficult due to unexpected ions and detector inaccuracies.
• Accurate annotation is crucial for identifying peptides and proteins in complex biological samples.

Purpose of the Study:

• To develop a deep convolutional neural network (CNN) for improved MS/MS spectrum annotation.
• To enhance the confidence and number of spectrum annotations using optimal feature extraction.
• To provide a faster and effective solution for low-resolution mass spectrometry data.

Main Methods:

• Developed Slider, a deep CNN that learns optimal feature extraction for MS/MS spectra.
• Trained and evaluated Slider on publicly available datasets.
• Compared Slider's performance against state-of-the-art methods like BoltzMatch, Res-EV, and Prosit.

Main Results:

• Slider achieved slightly higher spectrum annotation rates than existing methods.
• Slider operated 2-10 times faster than state-of-the-art comparison methods.
• Slider demonstrated high performance with low-resolution data, achieving nearly the same annotation rates as high-resolution methods.

Conclusions:

• Slider offers an effective approach for high-confidence MS/MS spectrum annotation.
• The method is particularly beneficial for researchers using older, low-resolution mass spectrometers.
• Slider provides a computationally efficient and accurate tool for peptide identification in proteomics.