Batch correction of single-cell sequencing data via an autoencoder architecture.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Technical variations in gene expression sequencing experiments introduce batch effect biases.
• These biases obscure true biological variations and hinder multi-dataset integration.
• Accurate joint analysis of single-cell sequencing data requires effective batch effect correction.

Purpose of the Study:

• To develop a novel method for integrating single-cell sequencing datasets by correcting batch effects.
• To preserve true biological variations while removing technical biases.
• To outperform existing state-of-the-art batch correction methods.

Main Methods:

• Developed Autoencoder-based Batch Correction (ABC), a semi-supervised deep learning architecture.
• Employed guided data compression with supervised cell type classifiers for biological signal retention.
• Utilized adversarial training for aligning different data batches.

Main Results:

• ABC effectively removes various types of batch effects from single-cell sequencing data.
• The method successfully preserves intricate biological variations.
• Comprehensive evaluations show ABC outperforms 10 state-of-the-art batch correction techniques.

Conclusions:

• ABC provides a robust solution for integrating single-cell sequencing datasets.
• The method enhances the reliability of biological conclusions drawn from multi-dataset analyses.
• ABC represents a significant advancement in computational approaches for single-cell genomics.