Single-cell RNA-seq interpretations using evolutionary multiobjective ensemble pruning.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers powerful insights into cellular heterogeneity.
• Existing computational methods for cell population identification face challenges like noise, high dimensionality, and scalability.
• There is a need for robust algorithms to accurately analyze scRNA-seq data and discover cell types.

Purpose of the Study:

• To develop a novel computational method for identifying cell populations from scRNA-seq data.
• To address limitations of existing methods, including experimental noise, numerical instability, high dimensionality, and computational scalability.
• To enhance the discovery of cell types and subtypes using evolutionary multiobjective optimization.

Main Methods:

• Proposed an evolutionary multiobjective ensemble pruning algorithm (EMEP).
• Applied unsupervised dimensionality reduction to create low-dimensional subspaces for clustering.
• Utilized cluster ensembles and dynamic selection of suitable clustering results.
• Formulated objective functions based on cluster validity indices (overall cluster deviation, within-cluster compactness, number of clusters) for evolutionary optimization.

Main Results:

• EMEP demonstrated superior performance compared to other clustering algorithms on 55 simulated and seven real scRNA-seq datasets.
• Successfully identified cell populations with clarity, including on a large-scale dataset (3005 cells, 4412 genes).
• Case studies revealed mechanistic insights into the biological relevance of identified cell populations.

Conclusions:

• EMEP effectively overcomes realistic restrictions in scRNA-seq data analysis.
• The algorithm provides a robust and scalable approach for cell type discovery.
• EMEP enhances the ability to identify and characterize cell populations from complex genomic data.