Identifying phenotype-associated subpopulations through LP_SGL.

Area of Science:

• Computational biology
• Genomics
• Immunology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity in diseases.
• Existing tools often overlook cell-cell interactions and group structures when identifying cell subpopulations.
• Identifying distinct cell subpopulations is vital for disease diagnosis and treatment response prediction.

Purpose of the Study:

• To develop a novel computational method, LP_SGL, that integrates cell group structures into subpopulation identification.
• To enhance the accuracy of identifying phenotype-associated subpopulations by incorporating scRNA-seq, bulk expression, and bulk phenotype data.
• To validate the clinical relevance of identified subpopulations and their associated signaling genes in predicting cancer, immunotherapy response, and survival.

Main Methods:

• Employed the Leiden algorithm to identify cell groups from scRNA-seq data, enhancing model robustness.
• Integrated scRNA-seq data with bulk expression and phenotype data using the proposed LP_SGL method.
• Compared LP_SGL performance against existing methods (Scissor, scAB) on lung adenocarcinoma, melanoma, and liver cancer datasets.

Main Results:

• LP_SGL identified a higher percentage of cancer cells, T cells, and tumor-associated cells compared to Scissor and scAB.
• Demonstrated superior performance of LP_SGL in lung adenocarcinoma diagnosis, melanoma drug response, and liver cancer survival prediction.
• Biological analysis revealed that signaling genes from identified cell subsets can predict cancer, immunotherapy response, and patient survival across multiple datasets.

Conclusions:

• LP_SGL effectively incorporates cell group structures to improve the identification of biologically and clinically relevant cell subpopulations.
• The method enhances the predictive power of scRNA-seq data for disease diagnosis, treatment response, and patient survival.
• Signaling genes within identified cell subsets represent promising biomarkers for cancer and immunotherapy outcomes.