A Systematic Evaluation of Supervised Machine Learning Algorithms for Cell Phenotype Classification Using Single-Cell...

Area of Science:

• Computational biology
• Genomics
• Machine learning

Background:

• Single-cell RNA sequencing (scRNA-seq) generates vast datasets for cell phenotype classification.
• Supervised machine learning models are increasingly used for scRNA-seq data analysis.
• Systematic performance evaluation of these algorithms across dataset sizes is lacking.

Approach:

• Evaluated 13 supervised machine learning algorithms on real and simulated scRNA-seq datasets of varying sizes.
• Assessed classification performance using metrics like AUC, F1-score, Precision, and Recall.
• Examined gene-selection performance on simulated datasets with known gene lists.

Key Points:

• ElasticNet with interactions performed best on small and medium datasets.
• XGBoost demonstrated excellent performance on large datasets.
• Linear Discriminant Analysis is the fastest method, suitable when computational speed is critical.

Conclusions:

• Algorithm choice for scRNA-seq cell phenotype classification depends on dataset size and speed requirements.
• ElasticNet and XGBoost are top performers for classification accuracy on different scales.
• Linear Discriminant Analysis provides a fast and scalable alternative without significant performance compromise.