Enhancing mass spectrometry imaging accessibility using convolutional autoencoders for deriving hypoxia-associated...

Area of Science:

• Oncology
• Biomedical Research
• Analytical Chemistry
• Data Science

Background:

• Mass spectrometry imaging (MSI) is underutilized for biomarker discovery in biomedical research despite its potential for studying cancer intratumoral heterogeneity.
• Challenges in MSI include high dimensionality, multicollinearity, and the direct output of mass-to-charge ratios rather than specific biochemical compounds.
• Tumor hypoxia is a key factor in cancer progression and exhibits significant spatial heterogeneity, making it a relevant target for MSI analysis.

Purpose of the Study:

• To develop a framework that enhances the accessibility of low-abundant signals in MSI data for biomarker discovery.
• To apply convolutional autoencoders to aggregate features associated with tumor hypoxia in cancer xenograft models.
• To improve the biological relevance of insights derived from MSI data for cancer research.

Main Methods:

• Utilized convolutional autoencoders (a type of deep learning model) to process and analyze high-dimensional MSI data.
• Focused on aggregating features related to tumor hypoxia, a spatially heterogeneous parameter.
• Employed ablation experiments to assess the relevance of individual hyperparameters and unraveled feature contributions.

Main Results:

• Demonstrated that MSI can capture low-abundant signals relevant to tumor hypoxia.
• Showed that convolutional autoencoders can effectively preserve these low-abundant signals within their latent space.
• Identified multiple hypoxia-associated peptide candidates by complementing MSI with tandem mass spectrometry (MS/MS) data.

Conclusions:

• The proposed autoencoder framework significantly improves the accessibility and preservation of low-abundant signals in MSI data.
• This approach yields more biologically relevant insights for cancer biomarker discovery compared to traditional methods like random forests alone.
• The study highlights the potential of integrating deep learning with MSI for advancing cancer research and identifying novel biomarkers.