scPER: A Rigorous Computational Approach to Determine Cellular Subtypes in Tumors Aligned With Cancer Phenotypes From...

Area of Science:

• Computational biology
• Genomics
• Immunology

Background:

• Single-cell RNA sequencing (scRNA-seq) offers insights into cellular diversity but faces challenges in processing large patient cohorts.
• Identifying phenotype-associated cell populations in bulk RNA sequencing (bulk RNA-seq) data is crucial for understanding tumor microenvironments.

Purpose of the Study:

• To develop a robust computational approach, scPER, for estimating cell compositions in bulk RNA-seq samples using scRNA-seq reference data.
• To enhance the identification of phenotype-associated cell subclusters and predict clinical outcomes, such as immunotherapy response.

Main Methods:

• scPER combines adversarial autoencoder and extreme gradient boosting for cell proportion estimation.
• It integrates diverse scRNA-seq datasets to build comprehensive reference panels and disentangle biological signals from technical confounders.
• The approach was validated against established methods like CIBERSORTx, BayesPrism, and Scaden.

Main Results:

• scPER demonstrated superior accuracy in cellular proportion estimation compared to existing methods.
• It accurately predicted immunotherapy response in metastatic melanoma and identified a novel T cell subcluster (FCRL3+, SLAMF7+).
• In metastatic urothelial cancer, scPER predicted TGFβ-mediated inhibition of CD4 naïve T cells impacting PD-L1 blockade efficacy.

Conclusions:

• scPER provides a robust method for integrating scRNA-seq data to estimate cellular proportions across various tumor types.
• The approach facilitates the identification of clinically relevant cell populations and subtypes, aiding in biomarker discovery and therapeutic strategy development.