Deep embedded clustering with multiple objectives on scRNA-seq data.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-throughput analysis of individual cell gene expression.
• Deep Embedded Clustering (DEC) shows promise for cell type identification in high-dimensional scRNA-seq data.
• Optimizing deep network architectures for DEC remains a challenge.

Purpose of the Study:

• To develop an automated method for optimizing DEC hyperparameters and architectures for scRNA-seq data.
• To enhance the accuracy and robustness of cell type identification using scRNA-seq data.
• To provide novel insights into cell type mechanisms through advanced clustering.

Main Methods:

• Integration of a denoising autoencoder for dimensionality reduction of scRNA-seq data.
• Implementation of an evolutionary multiobjective optimization framework for DEC.
• Formulation of three objective functions to balance model generality and clustering performance.
• Utilization of migration and mutation operators for hyperparameter and architecture optimization.

Main Results:

• The proposed evolutionary multiobjective DEC algorithm significantly outperforms existing state-of-the-art clustering methods on synthetic and real scRNA-seq datasets.
• Validation across multiple single-cell sequencing platforms demonstrates the algorithm's effectiveness.
• Experimental results show improved performance across various evaluation metrics.

Conclusions:

• The evolutionary multiobjective DEC provides an effective automated approach for cell type identification from scRNA-seq data.
• The method offers robust and accurate clustering, leading to novel biological insights.
• This approach advances the analysis of complex single-cell transcriptomic data.