Lagged principal trend analysis for longitudinal high‐dimensional data

Abstract

Biological systems are dynamic in nature, where time‐lagged correlations are commonly observed. Motivated by the problem of integrating longitudinal high‐dimensional datasets with time‐lagged temporal associations, we have developed a new statistical learning method named lagged principal trend analysis (LAPTA). Specifically, given longitudinal high‐dimensional datasets of two cohorts, LAPTA can extract time‐lagged associations and identify relevant features. The practical merits of LAPTA have been demonstrated through simulations and applications to host transcriptional response to influenza A infection data.