Multiscale Methods for Signal Selection in Single-Cell Data

Area of Science:

• Computational Biology
• Genomics
• Data Science

Background:

• Single-cell transcriptomics analysis commonly uses clustering and differential gene expression (DGE).
• These methods primarily identify discrete cell types and markers.
• Continuous biological variation within and between cell types may be missed.

Purpose of the Study:

• To develop unsupervised feature selection methods for single-cell transcriptomics.
• To analyze both discrete and continuous transcriptional patterns simultaneously.
• To identify genes with coherent expression patterns across multiple scales.

Main Methods:

• Eigenscores (eigi): Spectral decomposition of Laplacian graph for low-frequency patterning.
• Multiscale Laplacian score (MLS): Identifies relevant scales and coherently expressed genes.
• Persistent Rayleigh quotient (PRQ): Separates genes involved in bifurcation processes (e.g., pseudotime).

Main Results:

• Applied methods to published single-cell transcriptomics datasets.
• Validated previously identified marker genes.
• Discovered additional biologically meaningful genes with coherent expression patterns.
• Provided multidimensional gene rankings and visualizations of gene relationships.

Conclusions:

• The proposed topological methods offer a powerful alternative to traditional DGE analysis.
• These techniques capture complex gene expression patterns, including continuous variation.
• The methods enhance the understanding of cellular heterogeneity and dynamics.