Overview of the GLP workflow. GLP begins by taking the RNA expression matrix (genes x cells) as input. Step 1: The optimal span ( α ) for locally estimated scatterplot smoothing regression (LOESS) regression is selected based on the Bayesian Information Criterion(BIC). Step 2: A first-round LOESS regression is performed to model the relationship between the positive ratio and average expression. Step 3: Outliers are detected using Tukey’s biweight method, and a second-round LOESS regression is then conducted, excluding these outliers to improve robustness. Step 4: Residuals from the second-round regression are locally standardized. Step 5: Highly informative genes are identified based on each gene’s standardized residuals.
