BioDiscViz: A visualization support and consensus signature selector for BioDiscML results

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Machine learning (ML) algorithms excel at identifying complex patterns and biomarker signatures where traditional statistics falter.
• Current ML literature often lacks state-of-the-art methodologies for building non-overfitting models.
• Existing tools like BioDiscML address overfitting through multiple evaluation strategies but require enhanced visualization and flexibility.

Purpose of the Study:

• To develop an interactive visualization tool, BioDiscViz, to improve the usability and accessibility of BioDiscML outputs.
• To provide researchers with advanced visual support for investigating ML-derived biomarker signatures.
• To facilitate the extraction of consensus signatures from ML models.

Main Methods:

• Development of BioDiscViz, a visual interaction tool integrated with BioDiscML.
• Implementation of various visualizations including Principal Component Analysis (PCA) plots, UMAP, t-SNE, heatmaps, and boxplots.
• Inclusion of filtering mechanisms to support the extraction of consensus biomarker signatures.

Main Results:

• BioDiscViz offers comprehensive summaries, tables, and graphics for analyzing ML model results.
• The tool visualizes the best ML models and their correlated features effectively.
• BioDiscViz supports the identification of consensus biomarker signatures through interactive filtering.

Conclusions:

• BioDiscViz significantly enhances the investigation of ML results, making complex analyses more accessible.
• The tool democratizes ML applications in research by providing intuitive visualization capabilities.
• BioDiscViz fosters new opportunities in ML-driven research by broadening community access and understanding.