In this vignette, we show how to use lglasso package to estimate single-stage or double-stage networks from longitudinal data.
In section 1 and 2, we use simulated data to demonstrate how to use the two main function lglasso() and CVlglasso() respectively.
In section 3 and 4, we use the real antibody titer data from a large vaccine study to show that the how the analysis can lead to meaningful biological interpretations.
The function Simulate() for generating the simulated data can be found in file simulations.R in folder scripts of this repo. You can use it to play around to test the main functions. Unfortunately, the antibody real data is not publicly available. If you need these data for your research, please contact me at chowstat@gmail.com .
The algorithm is pretty stable, as you can see it converges nicely for both the simulated data and real data. However, based on my experience, the computational speed is not ideal yet. For the real data in which there are 83 features, it takes lglasso 2.2 minutes to converge for single network estimation and 5.7 minutes for double network estimation ( the results are listed in section 3). For function CVlglasso which has employed parallel method such as foreach package to speed up the computation, it is advised to run it on HPC where more cores and memory could be allocated.