Deep scSTAR: leveraging deep learning for the extraction and enhancement of phenotype-associated features from...

Area of Science:

• Computational Biology
• Immunogenomics
• Cancer Research

Background:

• Single-cell sequencing reveals cellular heterogeneity crucial for understanding disease.
• Extracting meaningful phenotype features is hindered by noise and batch effects.
• Identifying specific cell phenotypes is vital for disease mechanism and therapeutic response studies.

Purpose of the Study:

• To introduce Deep scSTAR (DscSTAR), a deep learning tool for enhancing phenotype-associated features in single-cell data.
• To demonstrate DscSTAR's utility in identifying disease-relevant cell populations.
• To improve the analysis of complex biological signals for disease insights.

Main Methods:

• Development of Deep scSTAR (DscSTAR), a deep learning algorithm.
• Application of DscSTAR to single-cell and spatial transcriptomics datasets.
• Analysis of phenotype-specific information to link cellular features to disease pathology and treatment response.

Main Results:

• DscSTAR identified HSP+ FKBP4+ CD8+ T cells associated with immune dysfunction and immunotherapy resistance in non-small cell lung cancer.
• Enhanced spatial transcriptomics analysis in renal cell carcinoma revealed immune suppressive interactions.
• Highlighted S100A12+ neutrophils and cancer-associated fibroblasts in hepatocellular carcinoma contributing to immune barriers and immunotherapy resistance.

Conclusions:

• DscSTAR effectively models and extracts phenotype-specific information from complex single-cell data.
• The tool advances understanding of disease mechanisms, immune cell roles, and therapy resistance.
• DscSTAR facilitates the discovery of novel cellular biomarkers and therapeutic targets.