DECODE: deep learning-based common deconvolution framework for various omics data

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Deconvolution algorithms estimate cell-type abundances from tissue data for cohort analysis.
• Current methods are limited to single-omics data, restricting generalizability and scalability.
• A universal framework for multi-omics deconvolution is needed.

Purpose of the Study:

• To present DECODE, a universal deconvolution framework for cell types and states.
• To enable seamless integration of diverse multiomics tissue datasets at the cellular level.
• To address the gap in metabolomics deconvolution.

Main Methods:

• Developed a universal deconvolution framework (DECODE).
• Applied DECODE to transcriptomic, proteomic, and metabolomic data.
• Integrated diverse multiomics tissue datasets.

Main Results:

• DECODE outperformed state-of-the-art methods across various omics data, donors, and conditions.
• Achieved high robustness in real-world scenarios with incomplete reference data.
• Successfully filled the gap in metabolomics deconvolution.

Conclusions:

• DECODE is a powerful tool for extending multiomics cohort data to the cellular level.
• The framework demonstrates broad applicability and superior performance.
• DECODE enhances cellular-level analysis of large-scale biological data.