Statistical analysis of proteomics data: A review on feature selection.

Area of Science:

• Biomedical research
• Proteomics
• Data Science

Background:

• The rise of '-omics' strategies necessitates a shift from deductive to data-driven inductive approaches in scientific research.
• Proteomics data analysis presents unique challenges due to high dimensionality and data sparsity, requiring effective data reduction techniques.
• The complexity of the proteome, including proteoforms, surpasses other '-omes', making robust data analysis strategies essential.

Purpose of the Study:

• To provide an overview of available methods for proteomics data analysis, with a specific focus on biomedical translational research.
• To offer guidance on selecting appropriate statistical procedures for data reduction, feature selection, cross-validation, and functional analysis of proteomics profiles.
• To address the lack of a standardized decision-making workflow in proteomics data analysis and prevent misinterpretation of results.

Main Methods:

• Review of statistical approaches for feature selection in proteomics data.
• Comparison of various data reduction strategies applicable to high-dimensional biological data.
• Discussion of cross-validation and functional analysis techniques for proteomics profiles.

Main Results:

• Identified various feature selection methods crucial for drawing functional proteomics signatures (e.g., for classification, diagnosis, prognosis).
• Highlighted the importance of careful selection of data reduction strategies due to the diverse range of available approaches and their dependence on input data and desired output.
• Presented suggestions for choosing standard statistical procedures for effective proteomics data analysis.

Conclusions:

• Standardized workflows for proteomics data analysis are currently lacking, increasing the risk of erroneous interpretation.
• This review aims to guide researchers toward suitable analysis pathways, minimizing mistakes in biomedical proteomics research.
• The judicious application of feature selection and data reduction techniques is critical for extracting meaningful biological insights from complex proteomic datasets.