LIDER: cell embedding based deep neural network classifier for supervised cell type identification.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates large datasets requiring efficient cell type identification.
• Current methods often rely on unsupervised clustering followed by manual annotation, which can be time-consuming and subjective.

Purpose of the Study:

• To develop a deep supervised learning method for automated cell type identification in scRNA-seq data.
• To introduce LIDER (celL embeddIng based Deep nEural netwoRk classifier) as a novel approach for this task.

Main Methods:

• LIDER utilizes a stacked denoising autoencoder to learn cell embeddings.
• A deep neural network classifier is employed for predicting cell types based on these embeddings.
• A tailored reconstructed loss function is incorporated into the autoencoder.

Main Results:

• LIDER demonstrates accurate cell type identification across eight diverse scRNA-seq datasets.
• Performance is comparable or superior to state-of-the-art methods.
• The method shows robustness to batch effects, a common challenge in scRNA-seq analysis.

Conclusions:

• Deep supervised learning, as implemented in LIDER, shows significant potential for automating cell type identification in scRNA-seq data.
• LIDER provides an accurate, robust, and efficient alternative to traditional methods.
• The availability of LIDER's code facilitates its adoption and further development.