massNet: integrated processing and classification of spatially resolved mass spectrometry data using deep learning...

Area of Science:

• Biomedical data analysis
• Computational pathology
• Metabolomics

Background:

• Mass spectrometry imaging (MSI) offers label-free biochemical insights crucial for disease diagnosis.
• The high dimensionality and complexity of MSI data present significant computational challenges for analysis.
• Current preprocessing methods like peak picking can prematurely influence downstream tissue classification and biological interpretation.

Purpose of the Study:

• To introduce massNet, a deep learning model designed for efficient and accurate analysis of mass spectrometry imaging data.
• To demonstrate massNet's capability to handle complex MSI data without prior preprocessing or peak picking.
• To improve computational scalability, nonlinearity, and speed in MSI data analysis.

Main Methods:

• Development of a novel deep learning architecture, massNet, for MSI data analysis.
• Application of massNet to classify MSI data from a mouse brain tumor model.
• Incorporation of automated methods for identifying predictive features and potential tumor-delineating peaks.

Main Results:

• MassNet achieved higher accuracy in classifying MSI data compared to support vector machines.
• The deep learning model demonstrated a substantial increase in computational speed.
• Automated feature learning and peak identification were successfully implemented within the massNet architecture.

Conclusions:

• MassNet offers a scalable, nonlinear, and fast approach to MSI data analysis.
• The model effectively classifies MSI data and identifies relevant features without traditional preprocessing steps.
• Deep learning presents a promising avenue for advancing MSI-based disease diagnosis and interpretation.