MOSCATO: a supervised approach for analyzing multi-Omic single-Cell data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic sequencing advances personalized medicine by generating multimodal single-cell data.
• Feature selection is crucial for analyzing complex biological datasets.
• Existing methods may not fully leverage multimodal single-cell information for clinical outcome association.

Purpose of the Study:

• Introduce MOSCATO, a novel technique for supervised feature selection.
• Identify clinically associated variables from multimodal single-cell datasets.
• Relate molecular features to patient outcomes.

Main Methods:

• Summarize single-cell data using tensor representations.
• Apply regularized tensor regression for feature selection.
• Evaluate feature sets for each 'omic' type.

Main Results:

• MOSCATO demonstrates favorable performance in selecting network features.
• The method is applicable to real-world multimodal single-cell data, such as CITE-seq.
• Simulations confirm the robustness of the approach.

Conclusions:

• MOSCATO provides a valuable analytical framework for multimodal single-cell data.
• The technique facilitates supervised feature selection linked to clinical outcomes.
• Future extensions can incorporate advanced statistical modeling and covariate adjustments.