scDILT: A Model-Based and Constrained Deep Learning Framework for Single-Cell Data Integration, Label Transferring,...

Area of Science:

• Single-cell genomics
• Computational biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution cellular analysis.
• Integrating diverse scRNA-seq datasets is crucial for comprehensive biological insights.
• Existing integration methods often fail to preserve original cell-type annotations when merging new data.

Purpose of the Study:

• To develop a novel computational tool for robust scRNA-seq data integration.
• To ensure that integrated datasets maintain the integrity of cell clusters from reference datasets.
• To provide a method that effectively removes batch effects while preserving biological signals.

Main Methods:

• Introduced scDILT, a tool employing a conditional autoencoder and deep embedding clustering.
• Utilized homogeneous constraints to maintain reference dataset clustering patterns.
• Employed heterogeneous constraints to map new cells to existing annotations.

Main Results:

• scDILT demonstrated superior performance in data integration compared to existing methods.
• Evaluations on simulated and real-world datasets confirmed scDILT's effectiveness.
• Successfully applied scDILT for integrating multi-omics single-cell datasets.

Conclusions:

• scDILT is a promising tool for integrating scRNA-seq data from various sources (batches, experiments, time points).
• The method effectively addresses batch effects while preserving crucial cell-type information.
• scDILT facilitates more accurate and comprehensive single-cell data analysis.