Adaptive Weight Sampling and Graph Transformer Neural Network Framework for Cell Type Annotation of Scrna-seq Data

Advancements in RNA-seq technology and bioinformatics have presented unprecedented opportunities for cell annotation. In the correlation analysis of single cells, cell type annotation is the most common computational task in the downstream specific task. Different cell types differ in morphology, function, or biochemical properties, and these differences determine the specific function and role of cells in the organism. Traditional methods for cell annotation have been limited by high costs in terms of manpower and resources, as well as small-scale annotation capabilities. To tackle this challenge, a cell annotation model leveraging adaptive weight sampling and graph transformer is proposed. This model utilizes scRNA-seq data, with gene interaction data being processed through adaptive weight sampling. Subsequently, SGFormer is employed to extract embedding features and aggregate information, leading to cell annotation based on the model’s aggregated insights. Through 5-fold cross-validation, our model demonstrated outstanding performance across 8 datasets, underscoring its potential as a reliable tool for cell annotation.