Predicting cell types with supervised contrastive learning on cells and their types.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution cellular expression profiling, advancing studies of cellular diversity and function.
• scRNA-seq data analysis faces challenges including multicollinearity, data imbalance, and batch effects.
• Accurate cell type annotation is crucial for interpreting scRNA-seq data, classifying cells by gene expression.

Purpose of the Study:

• To propose a novel method, SCLSC (Supervised Contrastive Learning for Single Cell), for accurate and efficient cell type annotation of scRNA-seq data.
• To leverage supervised contrastive learning on instance-type pairs for improved cell and cell type representation.
• To enhance the transfer of knowledge from annotated cells to feature representations, increasing training efficiency.

Main Methods:

• Developed SCLSC, a supervised contrastive learning framework for cell type annotation.
• Applied contrastive learning to instance-type pairs, clustering cells of the same type in an embedding space.
• Utilized knowledge transfer from annotated cells to improve feature representation for scRNA-seq data.

Main Results:

• SCLSC achieved superior accuracy in cell type prediction compared to five state-of-the-art methods.
• The method demonstrated robust performance in identifying cell types across different batch groups.
• SCLSC successfully discriminated previously unseen CD19+ B cell subtypes in a real-world immune cell dynamics study.

Conclusions:

• SCLSC offers a scalable and efficient approach for cell type annotation in scRNA-seq data.
• The method enhances understanding of cellular heterogeneity and dynamics, particularly in complex biological systems.
• SCLSC's ability to generalize to new cell subtypes highlights its potential for broad applications in single-cell analysis.