Inferring disease progression stages in single-cell transcriptomics using a weakly supervised deep learning approach.

Area of Science:

• Genomics
• Computational Biology
• Biomedical Data Science

Background:

• Single-cell sequencing of patient tissues is crucial for understanding human disease mechanisms.
• Cellular heterogeneity and varying pathological stages within patient samples complicate differential gene expression analysis.

Purpose of the Study:

• To develop a novel deep learning approach, scIDST (single-cell inference of Disease Stages), to address cellular heterogeneity in patient-derived tissues.
• To enable more accurate identification of disease-associated molecular features by inferring individual cell disease progression levels.

Main Methods:

• A weakly supervised deep learning framework was employed to infer disease progression levels for individual cells.
• The scIDST model was trained and validated on patient-derived tissue data.

Main Results:

• scIDST successfully inferred disease progression levels, enabling the detection of differential gene expression in disease-relevant genes within individual cells.
• These disease-specific patterns were not identifiable through traditional comparative analyses between patients and healthy donors.
• Pretrained scIDST models demonstrated applicability across multiple independent datasets, aiding in the identification of cells linked to disease risks and comorbidities.

Conclusions:

• scIDST provides a robust computational strategy for analyzing single-cell sequencing data from heterogeneous patient tissues.
• This approach enhances the identification of bona fide disease-associated molecular features and facilitates a deeper understanding of disease mechanisms at the single-cell level.