Analyzing omics data by pair-wise feature evaluation with horizontal and vertical comparisons.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Feature relationships are crucial for understanding complex biological data.
• Existing methods like k-top scoring pairs (k-TSP) primarily use horizontal comparisons.
• There is a need for methods that capture more comprehensive feature relationships.

Purpose of the Study:

• To propose a novel method, vertical and horizontal k-TSP (VH-k-TSP), for identifying discriminative feature pairs.
• To evaluate the effectiveness of VH-k-TSP in comparison to existing feature selection techniques.
• To apply VH-k-TSP to real-world biological datasets, including genomics and metabolomics.

Main Methods:

• Developed VH-k-TSP by incorporating both vertical and horizontal comparisons to assess feature pair discriminative ability.
• Compared VH-k-TSP against support vector machine-recursive feature elimination, relative simplicity-support vector machine, k-TSP, and M-k-TSP.
• Utilized nine public genomics datasets and a metabolomics dataset for liver disease, employing a one-to-one method for multi-class problems.

Main Results:

• VH-k-TSP demonstrated superior performance compared to the other four methods across most genomics datasets.
• In liver disease metabolomics data, VH-k-TSP achieved an accuracy of 88.11% ± 3.30% for discriminating between cirrhosis and hepatocellular carcinoma.
• This accuracy significantly surpassed the 77.39% ± 4.10% and 79.28% ± 3.73% achieved by k-TSP and M-k-TSP, respectively.

Conclusions:

• Combining vertical and horizontal comparisons provides a more effective approach to defining discriminative feature pairs.
• VH-k-TSP offers an improved feature selection strategy for complex biological data analysis.
• The method shows promise for applications in disease diagnosis and biomarker discovery.