Likelihood-based cell type classification using single-cell RNA-sequencing datatextYu, ChangcreatorWu, ZhijinAdvisorDe Vito, RobertaReaderBrown University. Department of Biostatisticssponsor2021, None p.born digitalThesis (Sc. M.)--Brown University, 2021thesesBackground: Single-cell RNA sequencing (scRNA-seq) quantifies the whole transcriptome on a cellular level which enables the exploration of tissue heterogeneity in biological samples. One common application in scRNA-seq is cell type identification. Different sequencing platforms may be associated with different characteristics, thus expression signatures identified in one platform may not be directly applicable to classify cells sequenced by another platform. Objectives: In this study, we aim to build a supervised cell classification tool with a focus on transferable classification between sequencing platforms. Methods: We used a likelihood-based approach that models the gene expression with parametric distributions. The gene and cell type parameters were estimated with training data, followed by model selection to allow similar cell types to share estimated parameters. The likelihood of a new cells belonging to each cell type was calculated for prediction. For cross platform prediction, the estimated parameters were corrected for systematic batch effect before used for prediction. Additionally, our method was compared with other published supervised classification tools. Results: Our method could correctly predict cell types in testing set data and across datasets and was among one of the best performing methods. Conclusions: We built a likelihood-based cell classification tool that could learn cell type labels from one dataset and transfer to another dataset that had no previous annotation.single cell RNA-sequencingBioinformaticsClassificationEnglishRPB20210607