SAVER-X

SAVER-X denoises single-cell RNA sequencing (scRNA-seq) data by integrating transfer learning, a deep autoencoder, and a Bayesian model to recover and refine gene expression signals from sparse and noisy measurements.

Key Features:

• Transfer Learning Across Datasets: Employs transfer learning to leverage information from multiple source scRNA-seq datasets, including different labs, experimental conditions, and species, to improve denoising of target datasets.
• Deep Autoencoder Integration: Uses a deep autoencoder architecture to capture complex gene-gene relationships and compress expression patterns for noise reduction.
• Bayesian Modeling: Integrates a Bayesian probabilistic model to refine inferred gene expression levels and estimate uncertainty.
• Accepted Input Formats: Accepts input data in .csv, .txt, and .rds file formats.

Scientific Applications:

• Cross-Laboratory Comparisons: Enables consistent analysis of scRNA-seq data generated under different experimental conditions or by distinct laboratories.
• Multi-Species Studies: Facilitates comparative studies by harmonizing gene expression data across species.
• Enhanced Downstream Analyses: Improves clustering, differential expression analysis, and trajectory inference by providing denoised expression matrices.

Methodology:

Applies transfer learning from existing datasets, employs a deep autoencoder to learn transferable gene-gene relationships, and uses a Bayesian model to refine denoised expression estimates and quantify uncertainty.