GSCI: A Generative and Sparse Compressed Sensing Imputation Framework for Single-Cell RNA-Sequencing Dropout Recovery

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) data present significant challenges due to inherent sparsity and high dimensionality.
• Dropout events in scRNA-seq data lead to numerous missing values, complicating downstream analyses and interpretation.
• Accurate imputation and dimensionality reduction are crucial for reliable analysis of scRNA-seq data.

Purpose of the Study:

• To develop an integrated framework for robust data imputation and dimensionality reduction in sparse scRNA-seq datasets.
• To enhance the reconstruction of gene expression patterns, including nonlinear dependencies, despite data sparsity.
• To improve the recovery of biologically critical genes and maintain global expression consistency.

Main Methods:

• A probabilistic modeling approach combined with sparse optimization techniques.
• Incorporation of sparsity constraints and heuristic search strategies to improve accuracy, especially for low-expression genes.
• Joint modeling of distributional structure and sparsity for reliable expression data restoration.

Main Results:

• The proposed framework effectively addresses data imputation and dimensionality reduction challenges in scRNA-seq data.
• Demonstrated superior performance over existing methods in reconstruction error and structural fidelity across diverse datasets.
• Successfully recovered biologically critical genes and maintained global expression consistency, outperforming baselines.

Conclusions:

• The integrated recovery framework offers a powerful solution for handling sparse and high-dimensional scRNA-seq data.
• Jointly modeling distributional structure and sparsity is key to reliable restoration of single-cell expression data.
• This method significantly advances the analysis of scRNA-seq data, enabling more accurate biological discoveries.