Investigation of Time Trajectories of Immune Infiltration Using the Functional Principal Component Analysis by Conditional Estimation Algorithm

• Activity Number: 554 - Novel Methods in Longitudinal Data Analysis
• Type: Contributed
• Date/Time: Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Minya Pu* and Xinlian Zhang and Jingjing Zou and Karen Messer
• Companies: Moores Cancer Center, University of California, San Diego and University of California, San Diego and University of California at San Diego (UCSD) and University of California, San Diego
• Keywords: Functional principal component analysis; Principal Analysis by Conditional Estimation (PACE)

Abstract

Time trajectories of major cell types in immune infiltration in a liver injury model were determined and compared using a sparse functional principal components analysis via PACE. The overall mean was extracted using a LOWESS curve and a covariance matrix was aggregated and smoothed via a local weighted bi-linear Gaussian kernel, and was then projected to a positive-semidefinite manifold. Bandwidths were chosen by generalized cross validation. The eigenfunctions of the smoothed matrix were extracted to form the smoothed functional principal components. The fPC scores for the individual curves were estimated and used to cluster the time trajectories into groups of cell types with related time profiles within each experimental condition, using a Gaussian mixture model and the EM algorithm. Comparing experimental conditions, the time trajectories clearly exhibited different patterns. The statistical significance of the overall difference between experimental conditions was tested using a Fourier series decomposition followed by a generalized linear model with the Fourier coefficients as predictors and group indicator as response.