Improved gene regulatory network inference from single cell data with dropout augmentation.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) data frequently exhibits "dropout" events, where transcript expression is missed, leading to zero-inflated data.
• This zero-inflation poses a significant challenge for downstream analyses, particularly for inferring gene regulatory networks (GRNs).

Purpose of the Study:

• To introduce Dropout Augmentation (DA), a novel regularization method to enhance resilience to zero inflation in scRNA-seq data.
• To present DAZZLE, a robust autoencoder-based model for GRN inference that incorporates the DA concept for improved stability and performance.

Main Methods:

• Dropout Augmentation (DA): A model regularization technique that augments data with synthetic dropout events to combat zero inflation.
• DAZZLE: A stabilized autoencoder-based structure equation model for GRN inference, leveraging DA for enhanced robustness.

Main Results:

• Benchmark experiments demonstrated that DAZZLE outperforms existing GRN inference methods in terms of performance and stability.
• DAZZLE effectively processed a large-scale longitudinal mouse microglia dataset, handling over 15,000 genes with minimal preprocessing.
• Dropout Augmentation proved to be a valuable approach for addressing zero inflation beyond simple imputation.

Conclusions:

• DAZZLE offers a robust and stable solution for GRN inference from single-cell RNA sequencing data, suitable for real-world applications.
• Dropout Augmentation is a promising technique with potential applications extending beyond GRN inference in single-cell data analysis.