scButterfly: a versatile single-cell cross-modality translation method via dual-aligned variational autoencoders

Area of Science:

• Computational Biology
• Genomics
• Single-cell Analysis

Background:

• Simultaneous multi-omics profiling in single cells reveals cellular heterogeneity but faces challenges with noisy data and high costs.
• Existing computational methods for cross-modality translation are limited in application and effectiveness.

Purpose of the Study:

• To develop a versatile computational method, scButterfly, for robust single-cell cross-modality data translation.
• To improve the analysis of cellular heterogeneity and molecular hierarchy in multi-omics data.

Main Methods:

• scButterfly utilizes dual-aligned variational autoencoders and data augmentation.
• The method is validated on multiple datasets to assess its performance.

Main Results:

• scButterfly demonstrates superior performance over baseline methods in preserving cellular heterogeneity.
• The method successfully translates datasets across various contexts and reveals cell type-specific biological insights.
• scButterfly shows broad applicability in integrative multi-omics analysis, data enhancement, and cell type annotation.

Conclusions:

• scButterfly offers a powerful and versatile solution for single-cell cross-modality translation.
• The method enhances multi-omics data analysis, enabling deeper biological insights and overcoming current technical limitations.