Multi-assignment clustering: Machine learning from a biological perspective

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Analyzing large-scale molecular data often involves clustering, assuming genes with similar expression profiles share functions.
• Biological complexity means proteins can have multiple functions, necessitating advanced clustering strategies for accurate interpretation.
• Current clustering techniques assign molecules to single groups, potentially oversimplifying complex biological relationships.

Purpose of the Study:

• To develop and implement a multi-assignment clustering (MAsC) approach for analyzing large-scale molecular data.
• To enhance the biological interpretability of transcriptomics data by allowing molecules to be assigned to multiple clusters.
• To improve the accuracy and biological relevance of pathway analysis from high-throughput molecular data.

Main Methods:

• Implemented a multi-assignment clustering (MAsC) algorithm, enabling molecules to be associated with multiple clusters.
• Applied MAsC to high-throughput transcriptomics data.
• Evaluated the impact of MAsC on downstream pathway analysis and noise reduction.

Main Results:

• MAsC increased the power of downstream pathway analysis on transcriptomics data.
• The method identified pathways with high biological relevance to experimental contexts.
• MAsC reduced noise by excluding genes with low correlation to any cluster, improving data clarity.

Conclusions:

• The multi-assignment clustering methodology facilitates the translation of large-scale molecular data into actionable biological knowledge.
• MAsC offers a more nuanced approach to clustering complex biological systems compared to traditional single-assignment methods.
• The MAsC R package is available for broader scientific use.