Using neural networks for reducing the dimensions of single-cell RNA-Seq data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-Seq) offers unprecedented resolution into cellular heterogeneity.
• scRNA-Seq analysis presents significant computational challenges in data clustering, cell identification, and functional state determination.
• Existing computational methods struggle to effectively address the complexities of scRNA-Seq data.

Purpose of the Study:

• To develop and evaluate a novel neural network (NN)-based method for scRNA-Seq data analysis and retrieval.
• To improve the accuracy of cell clustering and cell type/state inference from scRNA-Seq datasets.
• To provide a scalable solution for characterizing cells within heterogeneous scRNA-Seq samples.

Main Methods:

• Development and testing of various neural network architectures for scRNA-Seq data.
• Incorporation of prior biological knowledge into NN models.
• Utilizing NNs to generate reduced-dimension representations of single-cell expression data.
• Benchmarking against existing methods using independent scRNA-Seq datasets.

Main Results:

• The developed NN method demonstrates superior performance in correctly grouping cells across diverse experiments.
• The NN approach significantly improves the accuracy of inferring cell types and states by querying large databases.
• Performance gains were observed even for datasets not included in the NN model training phase.

Conclusions:

• Neural networks provide a powerful and effective framework for analyzing complex scRNA-Seq data.
• The NN method enhances the ability to characterize cell populations and understand cellular functions.
• A web server implementation facilitates database queries, empowering researchers to analyze heterogeneous scRNA-Seq samples.