Analyzing omics data by feature combinations based on kernel functions.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Understanding molecular feature combinations is crucial for disease diagnosis and prognosis.
• Existing methods often analyze feature cooperation using only fixed patterns, like linear combinations, limiting comprehensive analysis.
• Biosystems exhibit complex and varied feature interactions, necessitating advanced analytical approaches.

Purpose of the Study:

• To develop a new omics data analysis method, KF-λ-TSP, for comprehensive feature combination evaluation.
• To explore both linear and nonlinear feature combinations using kernel functions.
• To improve the identification of meaningful molecular interactions for disease mechanism studies.

Main Methods:

• Proposes KF-λ-TSP, a novel method utilizing kernel functions to study feature relationships in omics data.
• Employs multiple kernel functions to evaluate feature interactions from diverse perspectives, including nonlinear combinations.
• Constructs an ensemble classifier using top-scoring feature pairs identified by the method.

Main Results:

• KF-λ-TSP with multiple kernel functions outperforms single-kernel approaches by evaluating feature combinations from multiple views.
• The method demonstrates superior performance compared to TSP family algorithms and previous conversion-based strategies in most omics data analyses.
• KF-λ-TSP achieves comparable results to popular machine learning methods while utilizing fewer feature pairs.

Conclusions:

• KF-λ-TSP provides a more comprehensive evaluation of molecular combinations by measuring interactions from multiple perspectives.
• The method's ability to capture both linear and nonlinear interactions aids in mining information relevant to physiological and pathological changes.
• This approach can significantly advance the study of disease mechanisms and biomarker discovery in omics research.