scESI: evolutionary sparse imputation for single-cell transcriptomes from nearest neighbor cells

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data is prone to noise from the ubiquitous dropout problem, impacting gene expression profiles.
• This noise hinders accurate downstream analyses such as cell type classification and trajectory inference.

Purpose of the Study:

• To develop a novel algorithm, evolutionary sparse imputation (ESI), to address the dropout problem in scRNA-seq data.
• To improve the quality and reliability of scRNA-seq data by reducing noise and imputing missing gene expression values.

Main Methods:

• Constructed a sparse representation model for single-cell transcriptomes utilizing gene regulation relationships.
• Designed an optimization framework based on nondominated sorting genetics to solve the sparse model.
• Incorporated topological cell relationships and gene expression variability into an iterative global optimization search.

Main Results:

• The ESI algorithm learned a Pareto optimal cell-cell affinity matrix, effectively modeling sparse relationships.
• scESI demonstrated superior performance over benchmark methods on simulated datasets across various metrics.
• Applied to real scRNA-seq data, scESI enhanced cell type classification, visualization, trajectory reconstruction, and differential gene expression analysis.

Conclusions:

• ESI effectively reduces noise and improves data quality in scRNA-seq datasets.
• The algorithm facilitates more accurate biological insights, including marker gene expression trend recovery, new cell type discovery, and identification of regulatory pathways.