Removal of batch effects using distribution-matching residual networks

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Experimental data analysis is subject to measurement errors, including systematic and random components.
• Novel biological technologies like mass cytometry and single-cell RNA sequencing (scRNA-seq) are particularly susceptible to systematic errors.
• Uncorrected systematic errors can significantly impact statistical analyses.

Purpose of the Study:

• To develop a novel deep learning approach for the effective removal of systematic batch effects from biological data.
• To address the challenge of data calibration in high-throughput biological measurements.

Main Methods:

• A deep learning method utilizing a residual neural network.
• Training the network to minimize the Maximum Mean Discrepancy between distributions of data from different batches.
• Application to mass cytometry and single-cell RNA-seq datasets.

Main Results:

• Demonstrated effective attenuation of systematic batch effects.
• Successful application to diverse high-throughput biological datasets.
• Improved data quality for downstream statistical analysis.

Conclusions:

• The proposed deep learning method offers a robust solution for mitigating systematic errors in biological data.
• This approach enhances the reliability and accuracy of analyses derived from mass cytometry and scRNA-seq.
• Publicly available code and data facilitate reproducibility and further research.