Batch effects correction in scRNA-seq based on biological-noise decoupling autoencoder and central-cross loss

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Batch effects in single-cell RNA sequencing (scRNA-seq) data arise from experimental variations, masking true biological signals.
• Current batch correction methods often reduce dimensionality, risking the loss of rare cell populations.

Purpose of the Study:

• To develop a novel model, BDACL, for effective batch effect correction in scRNA-seq data.
• To address the limitation of existing methods that may lose rare cell types during batch correction.

Main Methods:

• BDACL utilizes a Biological-noise Decoupling Autoencoder (BDA) for data reconstruction and preliminary clustering.
• A similarity matrix and hierarchical clustering tree are constructed to analyze batch relationships.
• Central-cross Loss (CL) is introduced, combining cross-entropy and Central Loss for improved clustering and embedding consistency.

Main Results:

• BDACL effectively mitigates batch effects in an unsupervised manner.
• The model reconstructs data and merges clusters using a hierarchical approach.
• BDACL demonstrates superior performance by preserving rare cell types compared to existing methods.

Conclusions:

• BDACL offers a robust solution for batch effect correction in scRNA-seq analysis.
• The method enhances data consistency and interpretability while retaining valuable biological information, including rare cell populations.