Characterizing the cell identity in a heterogeneous tissue is essential to the in-depth understanding of this sample. Existing single-cell techniques (e.g., flow cytometry or in situ cell florescent imaging) allow us to do so using the high/low signal of a combination of multiple signature molecules or even of a single marker. Recent advance of single-cell RNA-seq technology profiles the entire transcriptome of individual cells. Using a few marker genes to characterize cell type in this new technique is less reliable due to the high noise level and the dynamic transcription behavior. Nonetheless, the "noisy" but high-throughput transcriptome profiles provide adequate information to reveal the cellular identity and to understand the detail of the molecular characteristics. In this chapter, we will demonstrate a new method that is based on the supervised learning of the single-cell transcriptome profiles of many different known cell types. We will demonstrate how this technique solves the cellular identity problem.
Keywords: Artificial neural network; Single-cell RNA-seq; Supervised classifier.