Fast Computational Recovery of Missing Features for Large-scale Biological Data

The lack of feature information is common in biological data and can seriously degrade the performance of existing data analysis methods. This chapter focuses on missing gene features in single-cell transcriptomics data. In the rapidly development of single-cell sequencing, the latest technological advances have made it possible to measure the intrinsic activity of single cells on a large scale, and enable to analyze the composition of cells within tissues with high precision. Based on this technology, many important biological structure identification methods have been proposed for the analysis of gene data. However, the missing genetic features have seriously hindered the full exploration of the internal information of biological data. For most of existing datasets, only about 20% of the genetic profiles can be effectively measured. Facing this problem, this chapter proposes deep recurrent autoencoder learning to achieve accurate and rapid imputation of missing gene expressions from millions of cell expression data.