A case-study on the influence of noise to log-gain principles for flux dynamic discovery

In this paper we show problems associated with the log-gain procedure [13] for determining flux-dynamics from time series by means of applying noise to the data sets. We illustrate this by first creating a set of flux functions and using these flux functions to derive a time series which we then apply Gaussian noise to [7]. This perturbed time series is then used in the log-gain procedure to determine flux-dynamics. The error from the two sets of flux functions is found to be extremely large for signal-to-noise ratios of less than about 25. To further illustrate the disparity in the results, we use these derived flux functions to discover new time series. We show that the log-gain procedure is very susceptible to noise, and that for it to be of practical use with data collect in vivo it must be made much more robust.