Abstract: We proposed a novel data completion algorithm based on the nonnegative matrix factorization model. Using the iterative traversal, the algorithm determined the best construction matrix and rank value, which solved the problem of missing value in single cell transcriptome sequencing data (RNA-seq), and avoided the influence of deficiency of the single cell sequencing depth on cell typing analysis. The experimental result on chronic myeloid leukemia data shows that after the missing values are recovered by the completion algorithm, the cell typing is more clear, which verifies the effectiveness of the proposed algorithm.

Key words: single cell RNA-seq data, missing element, nonnegative matrix factorization