PerSEveML: A Web-Based Tool to Identify Persistent Biomarker Structure for Rare Events Using Integrative Machine...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Omics datasets present computational challenges due to high dimensionality, large size, and non-linear structures, especially when analyzing rare events.
• Existing bioinformatics tools have limited integration of machine learning (ML) for understanding the biological underpinnings of rare events.

Approach:

• Introduced PerSEveML, an interactive, web-based platform utilizing crowd-sourced intelligence for rare event prediction and feature selection.
• The framework integrates multiple ML methods, providing evaluation metrics, entropy, and rank scores to assess individual method contributions and organize features.
• Visually categorizes input features into selected, unselected, and fluctuating groups to facilitate hypothesis generation.

Key Points:

• PerSEveML addresses the challenge of analyzing high-dimensional omics data with rare events.
• The tool employs an integrative machine learning approach to predict rare events and identify key biological features.
• It generates hypotheses by revealing persistent structures within selected and unselected features.

Conclusions:

• PerSEveML successfully generated valid hypotheses across three diverse, complex biological datasets with extremely rare events.
• The platform offers a comprehensive overview of integrative ML approaches for omics data analysis.
• It aids researchers in uncovering meaningful biological insights from challenging datasets.