Fusion of topological features by graph convolutional network for the classification of single-cell RNA sequencing data

Cells are the most basic unit of life. Organisms have trillions of cells, and these cells perform all their life activities. Although the structures in cells are all broadly similar, they are often not identical in type. Therefore, the correct classification of cells is a very meaningful task. Single-cell RNA sequencing technology is available to facilitate our study of cell classification. single-cell RNA sequencing data contains detailed cell expression information. However, these data are not well utilized in previous methods. In this paper, we fuse topological features through graph convolutional network (FTGCN) for the classification of single-cell RNA sequencing data. First, we construct the k-nearest neighbor (KNN) graph based on the gene expression levels between cells to obtain the similarity and construct connections between cells. Then, FTGCN learns the topology and the features between cells, and the finally fused topological features are used for the classification of cells. In the experiment, the comparison results with the baselines proved the effectiveness of FTGCN.