scCorr

scCorr improves accuracy of gene-gene correlation estimates in single-cell RNA-seq by reducing dropout-induced zero inflation through merging transcriptomically similar cells using a graph-based k-partitioning algorithm.

Key Features:

• Graph-based k-partitioning: Utilizes a graph-based k-partitioning algorithm to partition and merge transcriptomically similar cells to reduce zero inflation in scRNA-seq data.
• Reduction of dropout effects: Merges similar cell clusters to minimize abundant zero values (dropout) that bias gene-gene correlation estimates.
• Robust correlation detection: Detects coexpressed gene pairs with higher sensitivity than nonclustering methods, identifying 71 of 85 correlated gene pairs in PBMCs within a dataset of 100 clusters.
• Comparative performance: Demonstrates performance on par with three previously published methods across datasets.
• Downstream analysis support: Produces correlation estimates usable for network construction and gene–gene interaction studies and identified CD4+ T cells in PBMCs with ROC AUC of 0.96.

Scientific Applications:

• Gene coexpression analysis: Enables accurate detection of gene-gene coexpression patterns in single-cell transcriptomic data.
• Gene network and interaction studies: Supports construction of gene networks and gene–gene interaction analyses to study cellular functions and disease mechanisms.
• Cell-type identification in PBMCs: Facilitates identification of cell types such as CD4+ T cells in peripheral blood mononuclear cell datasets.

Methodology:

Applies a graph-based k-partitioning algorithm to partition and merge transcriptomically similar cells, reducing zero values (dropout) while preserving local cell community structures for improved gene-gene correlation estimation.