scNovel: a scalable deep learning-based network for novel rare cell discovery in single-cell transcriptomics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for biomedical research, enabling cell-type discovery.
• Existing auto-annotation tools are fast but struggle with identifying novel, rare cell types.
• Discovering rare cell types is essential but often time-consuming and requires domain expertise.

Purpose of the Study:

• To develop and validate scNovel, a deep learning-based neural network for novel rare cell discovery.
• To address the limitations of current tools in identifying rare cell populations from scRNA-seq data.
• To provide a scalable and accurate tool for high-throughput clinical data analysis.

Main Methods:

• Development of scNovel, a deep learning model specifically designed for novel rare cell detection.
• Evaluation of scNovel on diverse scRNA-seq datasets varying in scale, protocol, and data imbalance.
• Benchmarking against state-of-the-art novel cell detection models using AUROC performance metrics.

Main Results:

• scNovel significantly outperforms existing methods for novel cell detection, achieving superior AUROC scores (average >94%).
• Demonstrated scalability of scNovel on a million-scale dataset.
• Identified three potential novel macrophage subtypes in a COVID-19 dataset, with consistent expression patterns of COVID-related genes.

Conclusions:

• scNovel is a powerful and accurate tool for discovering novel rare cell types from scRNA-seq data.
• The model demonstrates high performance across various datasets and scales.
• scNovel shows promise as a valuable tool for analyzing high-throughput clinical data, particularly in identifying disease-specific cell subtypes.