NNICE: a deep quantile neural network algorithm for expression deconvolution.

Area of Science:

• Genomics
• Computational Biology
• Immunology

Background:

• Cell-type composition is crucial for understanding health status.
• Bulk gene expression, single-cell RNA sequencing, and deconvolution methods offer insights into cellular composition.
• Accurate cell type estimation from bulk data remains a challenge.

Purpose of the Study:

• To develop a novel computational method for estimating cell type abundance and uncertainty from bulk RNA-seq data.
• To introduce the Neural Network Immune Contexture Estimator (NNICE) model for automated deconvolution.
• To assess the performance of NNICE in recovering ground-truth cell type fractions.

Main Methods:

• Developed a deep learning and quantile regression-based method named NNICE.
• Applied NNICE to automatically deconvolve bulk RNA-seq data.
• Validated NNICE's ability to estimate cell type fractions on unseen data.

Main Results:

• NNICE successfully recovered ground-truth cell type fractions from both simulated and real bulk RNA-seq data.
• The model achieved high accuracy, with Pearson correlations of R=0.9 for both pseudo-bulk and real bulk gene expression data.
• NNICE demonstrated superior performance compared to existing baseline deconvolution methods across all cell types.

Conclusions:

• NNICE integrates statistical inference with deep learning for precise cell type deconvolution.
• The method provides accurate and interpretable estimations of cell type abundance from bulk gene expression.
• NNICE offers a powerful tool for analyzing cellular composition in various biological contexts.