scDeconv

scDeconv deconvolves bulk DNA methylation (DNAm) data by integrating single-cell RNA sequencing (scRNA-seq) and paired bulk RNA–DNAm datasets to infer relative cell-type abundances and identify cell-type-specific differential DNAm.

Key Features:

• Trans-Omics Integration: Integrates scRNA-seq with paired bulk RNA–DNAm data to overcome the scarcity of cell-type-specific DNAm references for deconvolution.
• Assumption of Similar Cell Compositions: Assumes that paired scRNA-seq and DNAm samples have similar cell compositions to enable transfer of cell-content information across modalities.
• Ensemble Model Training: Trains an ensemble model that fits cell contents with DNAm features while co-training and adjusting paired RNA deconvolution to improve cell-type abundance predictions.
• Broad Applicability: Applies the trained model to other bulk DNAm datasets for predicting relative cell-type abundances and can extend to other omics such as ATAC-seq when appropriate paired data are available.
• Cell-Type-Specific Differential Feature Detection: Provides functions to identify cell-type-specific inter-group differential features from bulk DNAm data.

Scientific Applications:

• Cell Composition Inference: Predicts cell-type abundances in heterogeneous tissues from bulk DNAm profiles.
• Epigenetic Research: Enables analysis of DNAm patterns across cell types within complex tissues to study epigenetic variation.
• Trans-Omics Studies: Supports integrative analyses combining transcriptomic and epigenomic layers to interpret interactions between scRNA-seq and DNAm data.

Methodology:

Integrates scRNA-seq with paired bulk RNA–DNAm data and trains an ensemble model using a co-training approach that fits cell contents with DNAm features while adjusting paired RNA deconvolution to transfer deconvolution information to bulk DNAm samples.