MOADE: a multimodal autoencoder for dissociating bulk multi-omics data.

Area of Science:

• Single cell biology
• Computational biology
• Bioinformatics

Background:

• Tissue complexity challenges cell mapping and tumor microenvironment analysis.
• Existing deconvolution methods primarily focus on transcriptomics, limiting applicability to other omics.
• Ambiguous cell markers and reference-to-target differences hinder cross-omic deconvolution.

Purpose of the Study:

• To develop a multimodal autoencoder pipeline (MOADE) for digital dissociation of bulk tissues.
• To enable joint prediction of personalized multi-omic profiles and cellular compositions.
• To overcome limitations of transcriptomic-focused deconvolution methods.

Main Methods:

• MOADE utilizes a multimodal autoencoder to link multi-dimensional features.
• It employs pseudo-bulk data constructed from internal non-transcriptomic references and external scRNA-seq data.
• The pipeline jointly predicts multi-omic profiles and cellular compositions.

Main Results:

• MOADE demonstrates superior generalizability and fidelity compared to nine other deconvolution pipelines.
• Performance was validated through rigorous simulation experiments.
• The method was successfully applied to real multi-omic data from diverse tissue types.

Conclusions:

• MOADE offers a robust solution for digital dissociation of bulk tissues across multiple omics.
• The pipeline enhances the analysis of cellular composition and multi-omic profiles.
• MOADE represents a significant advancement in computational biology for complex tissue analysis.