Biostatistics Weekly Seminar

Single-cell RNA sequencing (scRNA-seq) technology has been widely used in biological studies and various publicly accessible scRNA-seq databases are being built nowadays, but scRNA-seq data from different sources may not be directly merged to use due to universal batch effects. How to effectively remove the batches but preserve the true biological information of datasets becomes an important challenge in single-cell data analysis. The state-of-the-art methods typically utilize covariance or shared cell types to combine individual datasets, and which show efficacious in merging homogeneous cell types (the same cell type with very similar transcriptome across datasets) in many biological applications. However, these methods have defects when handling heterogeneous cell types, such as the same cell type at different developmental stages or in drug-treated conditions. The lack of a global standard scale in their mathematical models is one of the major reasons for biological signal distortion in their integration process. To overcome this problem, we introduce an algorithm Reference Principal Component Integration (RPCI), which employs a reference feature eigenvector to decompose all the individual datasets and establishes a calibrated frame for data integration. Compared with the results of other methods, the integrated data of RPCI reserve subtle information of heterogeneous cells with tiny transcriptomic difference, largely improving the sensitivity and reliability of extended analyses that are based on data integration, e.g., constructing cell trajectory and detecting differentially expressed genes.

Zoom Link to Follow
16 November 2022
1:30pm