BuDDI: Bulk Deconvolution with Domain Invariance to predict cell-type-specific perturbations from bulk.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell experiments offer high cellular resolution but face challenges like dissociation difficulties, cost, and limited tissue availability.
• Bulk experiments provide broader sample coverage but lack cellular specificity.
• A gap exists between deep cellular profiles of limited samples and broad bulk profiles across many samples.

Purpose of the Study:

• To develop a method, BuDDI (Bulk Deconvolution with Domain Invariance), that integrates bulk and single-cell RNA sequencing data.
• To infer cell-type-specific perturbation effects by leveraging domain adaptation techniques.
• To bridge the gap between limited single-cell data and extensive bulk data for comprehensive biological insights.

Main Methods:

• BuDDI employs domain adaptation within a variational autoencoder (VAE) framework.
• It learns independent latent spaces for cell type proportion, perturbation effect, experimental variability, and residual variability.
• Perturbation responses are simulated by independently composing these latent spaces.

Main Results:

• BuDDI successfully learned domain-invariant latent spaces on data with matched samples.
• The method accurately predicted cell-type-specific perturbation responses even without single-cell perturbed profiles during training.
• BuDDI outperformed comparative methods and baselines in predicting perturbation responses and sex-specific differences across complex experimental designs.

Conclusions:

• BuDDI effectively integrates bulk and single-cell RNA sequencing data to achieve single-cell resolution for studying perturbations.
• The method offers a scalable approach to combine large reference atlases with bulk-profiled disease or treatment signatures.
• BuDDI enhances the study of perturbations, sex differences, and other biological factors at a cellular level.