SCDRHA: A scRNA-Seq Data Dimensionality Reduction Algorithm Based on Hierarchical Autoencoder.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional data in single-cell RNA sequencing (scRNA-seq) presents challenges for analysis.
• Dropout events and zero-inflated data are significant hurdles in scRNA-seq studies.
• Effective dimensionality reduction is essential for scRNA-seq data visualization and clustering.

Purpose of the Study:

• To propose a novel dimensionality reduction algorithm for scRNA-seq data.
• To address the challenges of noise and zero-inflation in scRNA-seq datasets.
• To enhance the performance of scRNA-seq data visualization and cell clustering.

Main Methods:

• Development of SCDRHA, a scRNA-seq data dimensionality reduction algorithm.
• SCDRHA utilizes a hierarchical autoencoder architecture.
• The algorithm comprises a deep count autoencoder (DCA) for denoising and a graph autoencoder for low-dimensional projection.

Main Results:

• SCDRHA demonstrates superior performance in dimension reduction compared to state-of-the-art methods.
• The algorithm effectively reduces noise in scRNA-seq data.
• Experiments on five real scRNA-seq datasets show improved data visualization and cell clustering.

Conclusions:

• SCDRHA offers an effective solution for dimensionality reduction in scRNA-seq data.
• The proposed method successfully handles noise and zero-inflation.
• SCDRHA significantly enhances downstream analyses like cell clustering and visualization.