moSCminer: a cell subtype classification framework based on the attention neural network integrating the single-cell...

Area of Science:

• Computational Biology
• Genomics
• Epigenomics

Background:

• Precise cell subtype annotation is crucial for understanding cellular processes like differentiation.
• Current methods rely on manual, labor-intensive analysis of single-omics data, missing inter-omic correlations.
• Automated computational frameworks are needed for accurate cell subtype identification.

Purpose of the Study:

• To develop and validate moSCminer, a novel computational framework for cell subtype classification.
• To leverage single-cell multi-omics data for improved accuracy and biological insight.
• To provide an accessible, web-based platform for researchers.

Main Methods:

• Development of moSCminer, an attention-based neural network framework.
• Integration of three single-cell omics datasets: gene expression, DNA methylation, and DNA accessibility.
• Comparative performance evaluation against standard machine learning classifiers.

Main Results:

• moSCminer demonstrated superior performance in cell subtype classification compared to existing methods.
• The framework effectively learned the relative significance of different omics features.
• An omics-level attention module identified potential cell subtype markers.

Conclusions:

• Single-cell multi-omics integration significantly enhances cell subtype identification accuracy.
• moSCminer offers a robust and scalable solution for cell subtype classification.
• The study pioneers the integration of three single-cell omics datasets for this purpose.