Methods of Identifying Cell Type from Single Cell RNA-seq Data and the Interpretation

The ability to profile expression data in single cells (scRNA-Seq) has led to several important studies along with several new computational challenges, including questions about how to identify unique group of cells based on the expression profile. Different types of cells will have different gene expression profiles, thus they can be identified by these profiles. Predicting the cell type through expression profiles is of great practical significance. Previous application includes classifying cancers to specific diagnostic categories based on their gene expression, demonstrating the potential applications of these methods for tumor diagnosis and the identification of candidate targets for therapy. In this work, this task of cell type identification is pursued by implementing and comparing multiple machine learning methods. The neural network model can reach an accuracy of 59 percent when predicting 21 cell types. This work also interprets the neural network model by finding out the most important genes for classification through different methods that verify each other. Explanation and analysis based on prior biological knowledge are also included in this work.