Evaluating batch correction methods for image-based cell profiling.

Area of Science:

• Cellular imaging
• Computational biology
• Bioinformatics

Background:

• High-throughput image-based profiling generates vast cellular data for biological discovery.
• Batch effects from different labs and equipment limit data integration and interpretation.
• Standardized methods are needed to address batch variation in large-scale profiling studies.

Purpose of the Study:

• To benchmark single-cell RNA sequencing (scRNA-seq) batch correction techniques for image-based profiling data.
• To evaluate method performance across various complexity scenarios, including multi-lab and multi-microscope datasets.
• To provide a framework and metrics for assessing future batch correction algorithms.

Main Methods:

• Benchmarking ten scRNA-seq batch correction methods on the JUMP Cell Painting dataset.
• Testing performance across five scenarios of increasing batch effect complexity.
• Evaluating methods based on data integration quality and computational efficiency.

Main Results:

• Harmony and Seurat RPCA consistently ranked among the top three methods across all tested scenarios.
• Both top-performing methods demonstrated computational efficiency.
• The study established a robust framework and metrics for future batch correction assessments.

Conclusions:

• Harmony and Seurat RPCA are effective and efficient solutions for mitigating batch effects in image-based profiling data.
• The developed benchmarking framework will facilitate the advancement of data integration techniques.
• This work enhances the utility of public Cell Painting datasets for broader scientific discovery.