Topic model-based mass spectrometric data analysis in cancer biomarker discovery studies

Area of Science:

• Biomolecular analysis
• Computational biology
• Mass spectrometry

Background:

• Heterogeneity in human biospecimens presents a significant challenge for accurate biomolecule quantitation in cancer biomarker discovery.
• This issue, well-recognized in genomics, requires rigorous investigation in mass spectrometry-based proteomics and metabolomics.
• Effective purification of mass spectrometry data is crucial for reliable quantitative comparisons of biomolecules across samples.

Purpose of the Study:

• To investigate topic models for computational analysis of mass spectrometry data, addressing biospecimen heterogeneity.
• To evaluate the models' capability in inferring mixture proportions and identifying cancer profiles using both peak intensities and scan-level features.
• To assess the impact of data purification on cancer biomarker discovery in hepatocellular carcinoma (HCC) studies.

Main Methods:

• Employed topic models, including probabilistic generative models, to analyze mass spectrometry data (LC-MS and GC-MS).
• Incorporated both integrated peak intensities and scan-level features (extracted ion chromatograms) for comprehensive data analysis.
• Validated models using synthetic data and real-world datasets from HCC patients and liver cirrhosis controls.

Main Results:

• Topic models accurately inferred mixture proportions and underlying cancer profiles with low error (<7%) on synthetic data.
• Purification of mass spectrometry data identified more proteins and metabolites with significant changes between HCC and cirrhotic samples.
• Biomarkers selected post-purification improved pathway analysis and enhanced disease discrimination (increased AUC).

Conclusions:

• Topic model-based inference methods computationally address sample heterogeneity in LC/GC-MS analyses.
• Incorporating scan-level features enhances purification accuracy by minimizing information loss from peak integration.
• Topic model-based data purification offers significant benefits for mass spectrometry-based cancer biomarker discovery studies.