Inverse Weighting Method with Jackknife Variance Estimator for Differential Expression Analysis of Single-cell RNA Sequencing Data
Open AccessRecently developed single cell RNA sequencing (scRNA-seq) technology is able to measure the expression values of thousands of genes at the single cell level and has attracted substantial research from the computational and statistical community. However, scRNA-seq data have a high level of heterogeneity including extremely large outliers and an unusual abundance of observed zero counts due to biological stochasticity and technical difficulties. In this thesis, we review five cutting-edge analysis methods for differential expression analysis of scRNA-seq data: Limma, edgeR, MAST, scImpute and SAVER. Furthermore, an inverse non-dropout-probability weighting method is proposed to handle the informative dropout events in scRNA-seq data where the dropout probability depends on the gene-cell-specific expression level. The weights are estimated using the maximum likelihood method with the Gauss-Hermite quadrature. The variance of the weighted estimator is estimated using the jackknife method. Both normalized scRNA-seq data using linear regressions and negative binomial outcomes using generalized linear models are considered. Furthermore, the proposed method is extended to the lineage data where cells nested within the same embryo are correlated. Extensive simulation studies are carried out and our proposed method performs among the top three in all the methods in terms of AUC, sensitivity, specificity and FDR. Surprisingly, some computation-intensive methods such as scImpute and SAVER do not outperform the simpler methods when data are not robust when the data have a substantial amount of noise such as high-expression outliers, varying dispersion parameters generated from a Splat pipeline mimicking the real scRNA-seq data. One possible explanation is that those methods that take advantage of relationships between genes in the same cell or between similar cells are not robust when the data have a substantial amount of noise such as high-expression outliers, varying dispersion parameters(which depend on expected expression levels) and dropout probabilities (depending on true expression levels). Finally, our inverse weighting method and the test based on the jackknife variance estimator are illustrated using a lineage dataset where genes that are differentially expressed between cell division lineages are identified.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.