A comparison of methods for classifying clinical samples based on proteomics data: a case study for statistical and...

Area of Science:

• Biological Sciences
• Proteomics
• Bioinformatics

Background:

• Protein variation discovery is key for disease diagnosis.
• Omics disciplines generate high-dimensional data (p >> n).
• Many classification methods require dimensionality reduction for high-dimensional data.

Purpose of the Study:

• To compare statistical and machine learning approaches for proteomic data classification.
• To evaluate the utility of Partial Least Squares (PLS) and Principal Components Analysis (PCA) combined with classification methods.
• To contrast these with Support Vector Machines (SVM) for interpretability and performance.

Main Methods:

• Applied dimension reduction techniques: Partial Least Squares (PLS) and Principal Components Analysis (PCA).
• Utilized both statistical and machine learning classification methods post-dimension reduction.
• Compared performance against the Support Vector Machines (SVM) technique.

Main Results:

• Both PLS and SVM demonstrated strong utility in proteomic classification tasks.
• Statistical approaches offer interpretable classification rules, unlike 'black box' machine learning models.
• PLS proved effective for dimensionality reduction and classification in proteomic studies.

Conclusions:

• PLS and SVM are valuable tools for proteomic classification.
• Statistical methods offer advantages in result interpretability for disease diagnosis.
• Integrating dimension reduction with classification enhances the analysis of complex proteomic datasets.