CaSee: A lightning transfer-learning model directly used to discriminate cancer/normal cells from scRNA-seq

Abstract

Single-cell RNA sequencing (scRNA-seq) is one of the most efficient technologies for human tumor research. However, data analysis is still faced with some technical challenges, especially the difficulty in efficiently and accurately discriminate cancer/normal cells in the scRNA-seq expression matrix. In this study, we developed a cancer/normal cell discrimination pipeline called pan-cancer seeker (CaSee) devoted to scRNA-seq expression matrix, which is based on the traditional high-quality pan-cancer bulk sequencing data using transfer learning. It is compatible with mainstream sequencings technology platforms, 10x Genomics Chromium, Smart-seq2, and Microwell-seq. Here, CaSee pipeline exhibited excellent performance in the multicenter data evaluation of 11 retrospective cohorts and one independent dataset, with an average discrimination accuracy of 96.69%. In general, the development of a deep-learning based, pan-cancer cell discrimination model, CaSee, to distinguish cancer cells from normal cells will be compelling to researchers working in the genomics, cancer, and single-cell fields.

Anomaly pathway activities and transcriptome burden reveal cellular plasticity of human cancer

“…The framework of sceAPA consists of three major parts: a) Feature Extraction & Integration, which is in charge of converting 1D tensors into 2D tensors 22 (Fig. 1).…”