Sparsely-connected autoencoder (SCA) for single cell RNAseq data mining.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNAseq) is crucial for understanding cellular heterogeneity.
• Identifying functional characteristics of cell subpopulations from scRNAseq data presents a significant challenge.
• Existing methods may not fully capture the nuanced biological information within distinct cell clusters.

Purpose of the Study:

• To develop and present a novel computational tool for functional mining of single-cell clusters.
• To leverage Sparsely-Connected Autoencoder (SCA) for uncovering hidden biological features in scRNAseq data.
• To introduce new quality control metrics for evaluating clustering and model performance.

Main Methods:

• Development of a tool based on Sparsely-Connected Autoencoder (SCA) for functional analysis of single-cell clusters.
• Implementation of two novel metrics: Quality Control of Cluster (QCC) and Quality Control of Model (QCM).
• Integration of SCA with experimentally validated datasets including TF targets, miRNA targets, Kinase targets, and cancer-related immune signatures.

Main Results:

• SCA effectively uncovers hidden functional features specific to single-cell clusters.
• The encoded space derived from SCA can identify cluster-specific functional attributes.
• SCA's ability to reconstruct specific clusters highlights their importance in cell aggregation.
• QCC and QCM metrics provide quantitative evaluation of clustering and model performance.

Conclusions:

• The developed SCA-based tool enables effective functional mining of single-cell clusters from scRNAseq data.
• SCA facilitates the discovery of cell subpopulation-specific biological information.
• The tool, integrated into the rCASC framework with a GUI, simplifies complex analysis for researchers.