Digitaldlsorter

Digitaldlsorter performs deconvolution of bulk RNA-Seq expression profiles using Deep Neural Networks trained on single-cell RNA-Seq (scRNA-Seq) data to quantify cell-type proportions and distinguish compositional from cell-intrinsic expression changes.

Key Features:

• Deep Neural Network (DNN) model: Employs a DNN trained on scRNA-Seq datasets to perform deconvolution of bulk RNA-Seq data.
• Cell type quantification: Quantifies immune cell subpopulations including CD8+ T cells, CD4Tmem, CD4Th, CD4Tregs, B-cells, and estimates stromal content.
• Tumor-derived single-cell signatures: Preserves characteristics derived from scRNA-Seq of tumor tissues for application to bulk tumor samples.
• High-dimensional data utilization: Leverages the high dimensionality of single-cell data to overcome limitations of low-dimensional bulk RNA-Seq "skinny matrices."
• Accuracy and predictive power: Demonstrated high accuracy in quantifying cell populations and in predicting survival on synthetic bulk RNA-Seq data and TCGA samples.

Scientific Applications:

• Immune profiling: Discriminates specific immune cell populations within heterogeneous bulk samples to profile immune infiltration.
• Cancer microenvironment analysis: Applied to colorectal and breast cancer to characterize the tumor immune microenvironment from bulk RNA-Seq.
• Survival prediction and stratification: Enables survival outcome prediction and supports patient stratification analyses using deconvolution-derived cell-type estimates.
• Informing immunotherapy research: Provides cell-type-resolved insights that can inform personalized immunotherapy and patient-specific therapeutic analyses.

Methodology:

Trains a Deep Neural Network on single-cell RNA-Seq data to generate cell-type signatures that are applied to bulk RNA-Seq for deconvolution; analyses reported using R 3.51 and Python 3.6.8.