Fast and precise single-cell data analysis using a hierarchical autoencoder.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates large, noisy datasets, posing significant analytical challenges.
• Effective data processing is crucial for accurate biological insights from scRNA-seq.

Purpose of the Study:

• To introduce single-cell Decomposition using Hierarchical Autoencoder (scDHA), an advanced framework for scRNA-seq data analysis.
• To reliably extract representative cellular information and mitigate noise in scRNA-seq datasets.

Main Methods:

• scDHA employs a two-module pipeline: a non-negative kernel autoencoder for feature selection and a stacked Bayesian autoencoder for dimensionality reduction.
• The framework utilizes perturbation of the compressed space to prevent neural network overfitting and ensure generalized data representation.

Main Results:

• scDHA demonstrates superior performance compared to existing state-of-the-art methods in scRNA-seq data analysis.
• The framework excels in cell segregation, transcriptome visualization, cell classification, and pseudo-time inference tasks.

Conclusions:

• scDHA provides a robust and effective solution for analyzing complex scRNA-seq data.
• This framework significantly advances the capabilities for extracting meaningful biological information from single-cell transcriptomic studies.