hyperion

Overview

pharos runs NONMEM models from the command line. For modelers who work in R, that means switching to a terminal to check and run a model, then bringing the results back into R. Hyperion is the R interface to pharos, built on the same code, so a model is handled the same way from R or the command line. Reading in a model returns an object that shows its problem, dataset, run status, and initial parameter values. NM-TRAN can check it for control stream and dataset errors before it reaches the queue, and it can be submitted to a SLURM or SGE cluster from the same script. For every run, pharos records hashes of the model, dataset, and output files, so you can tell later whether any of them has changed. When the run finishes, its summary shows the minimization status, OFV, and condition number alongside the final estimates with standard errors and shrinkage. It flags problems such as an aborted covariance step or a parameter near a boundary, and the estimates, gradients, and shrinkage behind it read straight into data frames. Copying the model then starts the next one, optionally with its final estimates carried over as initial values. From reading a model to starting the next one, the whole loop fits in one R script. The results are already R objects: data frames for your own analysis, or parameter tables with hyperion.tables.