Development of a deep learning-based 1D convolutional neural network model for cross-species natural killer T cell identification using peripheral blood mononuclear cell single-cell RNA sequencing data.

Abstract

Background and aim: Natural killer T (NKT) cells exhibit the traits of both T and NK cells. Although their roles have been well studied in humans and mice, limited knowledge is available regarding their roles in dogs and pigs, which serve as models for human immunology. Single-cell RNA sequencing (scRNA-Seq) can elucidate NKT cell functions. However, identifying cells in mixed populations, like peripheral blood mononuclear cells (PBMCs) is challenging using this technique. This study presented the application of one-dimensional convolutional neural network (1DCNN) for the identification of NKT cells within scRNA-seq data derived from PBMCs.

Materials and methods: We used human scRNA-Seq data to train a 1DCNN model for cross-species identification of NKT cells in canine and porcine PBMC datasets. K-means clustering was used to isolate human NKT cells for training the 1DCNN model. The trained model predicted NKT cell subpopulations in PBMCs from all species. We performed Differential gene expression and Gene Ontology (GO) enrichment analyses to assess shared gene functions across species.

Results: We successfully trained the 1DCNN model on human scRNA-Seq data, achieving 99.3% accuracy, and successfully identified NKT cell candidates in human, canine, and porcine PBMC datasets using the model. Across species, these NKT cells shared 344 genes with significantly elevated expression (FDR ≤ 0.001). GO term enrichment analyses confirmed the association of these genes with the immunoactivity of NKT cells.

Conclusion: This study developed a 1DCNN model for cross-species NKT cell identification and identified conserved immune function genes. The approach has broad implications for identifying other cell types in comparative immunology, and future studies are needed to validate these findings.