A multi-label approach using binary relevance and decision trees applied to functional genomics.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Multi-label classification problems are common in bioinformatics, where data instances can belong to multiple categories simultaneously.
• Existing methods often prioritize performance over model interpretability, hindering biological insights.
• The Binary Relevance algorithm is a common baseline but does not inherently account for label dependencies.

Purpose of the Study:

• To propose a novel adaptation of the Binary Relevance algorithm for multi-label classification.
• To enhance the interpretability of multi-label classification models without sacrificing performance.
• To evaluate the proposed method on functional genomic datasets.

Main Methods:

• Developed a modified Binary Relevance algorithm incorporating label relationships.
• Conducted comparative experiments against established multi-label classification methods.
• Applied and validated the approach using real-world functional genomic datasets.

Main Results:

• The proposed method achieved performance comparable to existing state-of-the-art techniques.
• The adapted algorithm successfully provided an interpretable model for multi-label classification tasks.
• Experimental results demonstrated the efficacy of the approach on functional genomic data.

Conclusions:

• The novel Binary Relevance adaptation offers a balance between predictive performance and model interpretability.
• This approach is valuable for analyzing complex biological data with multiple functional annotations.
• The method facilitates a deeper understanding of biological systems through interpretable machine learning models.