A hybrid deep clustering approach for robust cell type profiling using single-cell RNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity but is prone to experimental noise and biological variation.
• Existing machine learning methods struggle to effectively handle noise and variation in scRNA-seq data.
• Accurate analysis of scRNA-seq data is crucial for discovering cell subtypes and states.

Purpose of the Study:

• To develop a robust machine learning approach for analyzing noisy scRNA-seq data.
• To improve the identification of cellular subtypes and states from transcriptomic data.
• To enhance cancer subclone identification and the study of clonal heterogeneity.

Main Methods:

• Developed deep unsupervised single-cell clustering (DUSC), a hybrid approach integrating deep learning for feature generation and model-based clustering.
• Implemented a novel technique to estimate the optimal number of latent features within the deep learning model.
• Applied DUSC to triple-negative breast cancer scRNA-seq data.

Main Results:

• DUSC effectively handles noise and variation in scRNA-seq data, producing compact and informative representations.
• The method achieves performance comparable to supervised learning, outperforming existing feature learning and clustering techniques.
• Identified potential cancer subclones in triple-negative breast cancer, highlighting the role of copy-number variation and clonal heterogeneity.

Conclusions:

• DUSC offers a robust and accurate method for analyzing scRNA-seq data, overcoming limitations of current approaches.
• The developed technique for estimating latent features enhances the reliability of deep learning models in bioinformatics.
• DUSC has the potential to advance our understanding of cellular atlases and improve patient diagnostics and treatment in oncology.