multiDGD: A versatile deep generative model for multi-omics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell genomics enables joint profiling of gene expression and chromatin accessibility.
• Increasing complexity of multi-omics data necessitates scalable integration methods.
• Existing models often lack functionality or scalability for multi-modal data analysis.

Purpose of the Study:

• Introduce multiDGD, a scalable deep generative model for multi-modal single-cell data integration.
• Provide a probabilistic framework for learning shared representations of transcriptome and chromatin accessibility.
• Facilitate downstream analyses such as data integration and association detection.

Main Methods:

• Developed a scalable deep generative model (multiDGD).
• Utilized a probabilistic framework for learning joint representations.
• Incorporated probabilistic modeling of sample covariates for post-hoc integration.
• Applied the model to human and mouse single-cell datasets.

Main Results:

• multiDGD demonstrates outstanding performance in data reconstruction without feature selection.
• The model learns well-clustered joint representations of transcriptome and chromatin accessibility.
• Probabilistic covariate modeling allows post-hoc data integration without fine-tuning.
• multiDGD effectively detects statistical associations between genes and regulatory regions.

Conclusions:

• multiDGD offers a scalable and effective solution for integrating multi-modal single-cell genomics data.
• The probabilistic framework facilitates robust data integration and downstream analysis.
• The model advances the analysis of gene expression and chromatin accessibility relationships.