Evaluation of single-cell RNAseq labelling algorithms using cancer datasets.

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for dissecting tissue cellular composition.
• Automated cell type identification methods are essential for analyzing complex biological systems like cancer.
• Cancer's tumor microenvironment presents challenges due to diverse, functionally distinct cell subpopulations.

Purpose of the Study:

• To comprehensively assess the performance of cell-based and cluster-based scRNA-seq labelling algorithms in a cancer-specific context.
• To compare the efficiency and accuracy of various automated cell type identification methods using cancer datasets.
• To identify optimal algorithms for clinical applications, including handling small or under-represented cell populations.

Main Methods:

• Evaluated 17 cell-based and 9 cluster-based scRNA-seq labelling algorithms.
• Utilized 8 diverse cancer datasets for large-scale performance assessment.
• Employed multiple performance metrics to quantitatively compare algorithms.
• Tested algorithm robustness in cross-cohort validation scenarios.

Main Results:

• Cell-based methods generally demonstrated higher performance and faster execution times than cluster-based methods.
• Cluster-based methods were more effective at labelling non-malignant cells, potentially due to limited malignant cell signatures.
• Larger training cell populations positively influenced prediction accuracy for cell-based methods.
• scPred and SVM algorithms showed superior overall performance on cancer-specific data.

Conclusions:

• Cell-based algorithms, particularly scPred and SVM, are recommended for automated cell type labelling in cancer scRNA-seq data.
• Algorithm selection should consider factors like cell type representation and clinical applicability.
• The developed analysis pipeline aids in evaluating and selecting appropriate scRNA-seq labelling tools.