Biomarker selection and sample prediction for multi-category disease on MALDI-TOF data

Area of Science:

• Biomedical data analysis
• Computational biology
• Proteomics

Background:

• Diseases progress through stages, necessitating stage-specific biomarkers.
• Accurate classification and biomarker identification are crucial for multi-category disease problems.
• Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) is a key technology in biomarker discovery.

Purpose of the Study:

• To develop and evaluate integrated classification and feature selection methods for disease staging.
• To identify reliable biomarkers for distinguishing between hepatocellular carcinoma (HCC), cirrhosis, and healthy states.
• To enhance the accuracy of disease classification using combined computational approaches.

Main Methods:

• A novel classification method integrating error-correcting output coding (ECOC) and pairwise coupling (PWC).
• An extended Markov blanket (EMB) feature selection method for identifying significant biomarkers.
• Analysis of a liver cancer MALDI-TOF MS dataset including HCC, cirrhosis, and healthy spectra.

Main Results:

• Discovery of distinct peak patterns for differentiating pairwise categories among HCC, cirrhosis, and healthy samples.
• Quantification of peak importance and reliability using weight values and frequencies.
• Demonstration of superior classification capability compared to classical ECOC, random forest, Naive Bayes, and J48 methods.

Conclusions:

• The proposed integrated approach effectively classifies disease stages and selects relevant biomarkers.
• The identified spectral patterns and biomarkers hold potential for early detection and diagnosis of liver diseases.
• This study provides a robust framework for biomarker discovery in complex diseases.