ZiPo: A Deep Neural Network to De-Noise Single-Cell RNA Sequencing Data

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity and dynamics.
• scRNA-seq data analysis faces challenges due to 'dropouts,' where genes show zero expression inaccurately.
• Existing methods struggle to fully capture or correct for these zero-inflation issues.

Purpose of the Study:

• To introduce ZiPo, a novel deep artificial neural network designed for scRNA-seq data analysis.
• To address the problem of measurement dropouts by incorporating adjustable zero inflation.
• To improve rate estimation and library size prediction for more accurate transcriptome analysis.

Main Methods:

• ZiPo utilizes deep autoencoders and Poisson/negative binomial distributions.
• Novel strategies include library size prediction and residual connections for enhanced performance.
• A scale-invariant loss term promotes model sparsity and biological interpretability.

Main Results:

• ZiPo effectively captures dropouts in scRNA-seq data.
• The model demonstrates superior performance compared to existing techniques across multiple datasets.
• Processing time is efficient, scaling linearly with the number of cells.

Conclusions:

• ZiPo offers a powerful and interpretable solution for scRNA-seq data analysis, particularly for handling dropouts.
• The method enhances the discovery of cellular populations and transcriptional dynamics.
• ZiPo provides a valuable tool for researchers in genomics and computational biology.