A new data analysis method based on feature linear combination.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Biology

Background:

• Biological data features exhibit complex relationships reflecting physiological and pathological states.
• Effective classification rules are crucial for disease discrimination and understanding disease mechanisms.
• Existing methods like k-top scoring pairs (k-TSP) explore feature relationships but can be improved for efficiency.

Purpose of the Study:

• To introduce a novel data analysis method, linear combination of k top scoring pairs (LC-k-TSP), for more efficient classification rule definition.
• To enhance the analysis of feature relationships in biological data for improved discrimination.
• To provide a method that facilitates biomedical explanation of classification rules.

Main Methods:

• LC-k-TSP utilizes Support Vector Machine (SVM) to determine the optimal linear combination for each feature pair.
• Feature pairs are scored based on the discriminative power of their linear combinations.
• An ensemble classifier is constructed using k disjoint top scoring pairs.

Main Results:

• LC-k-TSP demonstrated superior performance over k-TSP across twelve public datasets.
• The method achieved comparable accuracy rates to SVM and Random Forest (RF).
• Application to hepatocellular carcinoma (HCC) metabolomics data showed improved accuracy in distinguishing HCC and cirrhosis (CIR) groups compared to k-TSP.

Conclusions:

• LC-k-TSP offers a more effective approach to defining classification rules from biological data by considering unique linear relationships for each feature pair.
• The method provides simple, interpretable rules beneficial for biomedical exploration.
• LC-k-TSP shows significant potential for analyzing complex biological datasets and aiding in disease diagnosis.