CellScout: Visual Analytics for Mining Biomarkers in Cell State Discovery

Area of Science:

• Computational Biology
• Biomedical Informatics
• Systems Biology

Background:

• Cell state discovery is vital for understanding biological systems and medical advancements.
• Identifying cell-specific biomarkers is challenging due to the co-discovery process and visualization limitations.
• Current methods often rely on visual clustering, which can be inaccurate and lead to trial-and-error biomarker identification.

Purpose of the Study:

• To develop an effective computational tool for uncovering hidden associations between cell populations and biomarkers.
• To assist biologists in refining cell state discovery by exploring and validating biomarker relationships.
• To address the limitations of traditional dimensionality reduction and visual clustering in cell state analysis.

Main Methods:

• Designed a machine-learning algorithm utilizing the Mixture-of-Experts (MoE) technique.
• Developed a collaborative visual analytics system named CellScout.
• Validated the system through expert interviews and case studies.

Main Results:

• The Mixture-of-Experts algorithm successfully identified meaningful associations between cell populations and biomarkers.
• The CellScout system facilitated exploration and refinement of these association relationships.
• Case studies demonstrated the system's effectiveness in discovering novel cell states.

Conclusions:

• The developed machine learning algorithm and CellScout visual analytics system offer a robust solution for cell state and biomarker co-discovery.
• This approach enhances the accuracy and efficiency of identifying distinct cell populations and their defining biomarkers.
• The tool empowers biologists to advance cell state discovery, leading to better understanding of biological systems and improved medical outcomes.