Infusing structural assumptions into dimensionality reduction for single-cell RNA sequencing data to identify small...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables exploration of cellular heterogeneity.
• Existing dimensionality reduction methods are primarily data-driven.
• Incorporating biological assumptions can enhance scRNA-seq data analysis.

Purpose of the Study:

• To develop a novel dimensionality reduction approach for scRNA-seq data.
• To integrate unsupervised deep learning with boosting for formalizing biological assumptions.
• To identify key gene sets driving biological variation in scRNA-seq data.

Main Methods:

• Proposed the boosting autoencoder (BAE) method.
• Combined deep learning (autoencoders) with boosting algorithms.
• BAE selects small gene sets that explain latent dimensions.

Main Results:

• Demonstrated BAE's ability to perform dimensionality reduction on scRNA-seq data.
• Successfully identified biologically relevant gene sets associated with latent dimensions.
• Applied BAE to explore neural cell diversity and embryonic development patterns.

Conclusions:

• BAE offers a powerful framework for integrating biological knowledge into dimensionality reduction.
• The approach facilitates the exploration of cellular heterogeneity and biological processes.
• BAE enhances the interpretability of scRNA-seq data analysis.