SIMS: A deep-learning label transfer tool for single-cell RNA sequencing analysis.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Automated cell labeling is crucial for interpreting cell atlases.
• Current classification algorithms lack accuracy for single-cell RNA data.
• Efficient and interpretable tools are needed for single-cell analysis.

Purpose of the Study:

• Introduce SIMS (scalable, interpretable machine learning for single cell), a novel low-code pipeline for single-cell RNA classification.
• Evaluate SIMS's accuracy and robustness across diverse biological datasets.
• Demonstrate SIMS's utility in understanding cellular differentiation and identifying variations in organoid models.

Main Methods:

• Developed a data-efficient, low-code machine learning pipeline (SIMS) for single-cell RNA classification.
• Benchmarked SIMS using single-cell RNA datasets from various tissues and species.
• Applied SIMS to analyze neuronal subtypes in the developing brain and cell lines in human cortical organoids.

Main Results:

• SIMS achieves high accuracy in classifying cells, even with small training sets (<3,500 cells).
• The pipeline effectively predicts neuronal subtypes and identifies genetic changes during differentiation.
• SIMS detects cell-line specific differences and misannotated lineages in human cortical organoids.

Conclusions:

• SIMS is a versatile and robust tool for accurate cell-type classification from single-cell RNA data.
• The pipeline facilitates the study of cellular heterogeneity and differentiation processes.
• SIMS aids in quality control and comparative analysis of single-cell datasets, including organoid models.