scInterpreter: a knowledge-regularized generative model for interpretably integrating scRNA-seq data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data offers vast potential for biological discovery through integration analyses.
• Current integration models often lack interpretability and are challenging to train, limiting their widespread application.

Purpose of the Study:

• To develop an interpretable and efficient deep learning model for scRNA-seq data integration.
• To address the limitations of existing black-box integration methods.

Main Methods:

• Proposed scInterpreter, a novel deep learning-based interpretable model for scRNA-seq data integration.
• Evaluated scInterpreter's performance against state-of-the-art models on benchmark datasets.
• Assessed the model's extensibility for integrating and annotating atlas scRNA-seq data.
• Investigated the impact of knowledge priors on accelerating the training process.
• Performed interpretability analysis on cell representations in the embedding space.

Main Results:

• scInterpreter significantly outperformed existing state-of-the-art models in multiple benchmark datasets.
• The model demonstrated robustness across various evaluation scenarios.
• Incorporating a knowledge prior notably accelerated the training process.
• Interpretability analysis revealed that cell representations captured significant biological meaning, identifying novel genes associated with specific pathways in PBMC data.

Conclusions:

• scInterpreter is an effective and interpretable tool for scRNA-seq data integration.
• The model's cell representations are biologically meaningful, aiding in pathway and gene discovery.
• scInterpreter is expected to greatly facilitate single-cell transcriptomics research.