UCLA

This thesis focuses on applying the quantile generalized additive model (QGAM) to detect differentially expressed (DE) genes in the single-cell RNA sequencing data. Most of the existing DE gene inference methods are developed based on Generalized Additive Model (GAM). At the same time, GAM can be sensitive to outliers during the model fitting process. Such sensitivity impacts the accuracy of the DE gene detection for the dataset with outliers. We want to use QGAM’s robustness to outliers to improve the DE gene detection accuracy. We compared the performance of the QGAM-based DE gene inference method (qgamDE) with two state-of-the-art GAM-based methods, PseudotimeDE, and tradeSeq, by applying them to the simulated data. In conclusion, the performance of qgamDE and PseudotimeDE is better and more stable compared to the performance of tradeSeq. As a result, we added qgamDE to the PseudotimeDE R package as new functionality.