Prediction of distinct populations of innate lymphoid cells by transcriptional profiles

Abstract

Innate lymphoid cells (ILCs) are a unique type of lymphocyte that differ from adaptive lymphocytes in that they lack antigen receptors, which primarily reside in tissues and are closely associated with fibers. Despite their plasticity and heterogeneity, identifying ILCs in peripheral blood can be difficult due to their small numbers. Accurately and rapidly identifying ILCs is critical for studying homeostasis and inflammation. To address this challenge, we collect single-cell RNA-seq data from 647 patients, including 26,087 transcripts. Background screening, Lasso analysis, and principal component analysis (PCA) are used to select features. Finally, we employ a deep neural network to classify lymphocytes. Our method achieved the highest accuracy compared to other approaches. Furthermore, we identified four genes that play a vital role in lymphocyte development. Adding these gene transcripts into model, we were able to increase the model's AUC. In summary, our study demonstrates the effectiveness of using single-cell transcriptomic analysis combined with machine learning techniques to accurately identify congenital lymphoid cells and advance our understanding of their development and function in the body.

Keywords: DNN; LASSO; gene expression; innate lymphoid cells; machine learning.

FIGURE 1

The framework of our work.

FIGURE 2

Comparison of DNN method with several other models by AUC.

FIGURE 3

The confusion matrix of four models.

FIGURE 4

Model performance with or without four key genes.