scIAE: an integrative autoencoder-based ensemble classification framework for single-cell RNA-seq data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed analysis of cellular heterogeneity.
• Accurate cell type identification is crucial for building cell atlases and downstream analyses.
• scRNA-seq data present challenges due to high dimensionality, sparsity, and dropouts.

Purpose of the Study:

• To develop an effective and robust computational framework for cell type classification and disease status prediction using scRNA-seq data.
• To address the challenges of high-dimensional and sparse single-cell gene expression data.
• To create a flexible and adaptable model for diverse scRNA-seq applications.

Main Methods:

• Proposed scIAE, an integrative autoencoder-based ensemble classification framework.
• Employed multiple random projections and combined stacked, denoising, and sparse autoencoders for data representation.
• Built base classifiers on compressed representations and integrated their predictions.

Main Results:

• scIAE demonstrated effectiveness and robustness in feature extraction, independent of dimension choice.
• Achieved high classification power for cell type annotation (intradataset, across batches, platforms, and species).
• Successfully predicted disease status using scRNA-seq data.

Conclusions:

• scIAE provides a powerful and flexible tool for scRNA-seq data analysis.
• The framework enhances cell type annotation and disease prediction capabilities.
• scIAE is a valuable contribution to the field of single-cell genomics and precision medicine.