MSMCE: A novel representation module for classification of raw mass spectrometry data

Area of Science:

• Biomedical data analysis
• Computational biology
• Spectrometry techniques

Background:

• Mass spectrometry (MS) is vital in biomedicine but presents challenges due to high-dimensional and complex data.
• Current deep learning methods for MS classification have limited feature representation, often using single-channel data that misses structural information.

Purpose of the Study:

• To develop a novel module for generating effective multi-channel representations of raw mass spectrometry data.
• To improve feature extraction and classification performance in mass spectrometry data analysis.

Main Methods:

• Proposed a Multi-Channel Embedding Representation Module (MSMCE) to model inter-channel dependencies for raw MS data.
• Implemented a feature fusion mechanism by concatenating initial encoded representations with multi-channel embeddings.
• Evaluated the module on four public mass spectrometry datasets.

Main Results:

• The MSMCE module significantly improved classification performance on raw MS data.
• Demonstrated enhanced computational efficiency and training stability compared to existing methods.
• Effectively captured structural information within MS data through multi-channel representations.

Conclusions:

• The proposed MSMCE module offers a powerful approach for raw mass spectrometry data classification.
• MSMCE enhances classification accuracy, efficiency, and stability, showing potential for diverse biomedical applications.
• Multi-channel representations are key to overcoming limitations in current MS data analysis techniques.