Inferelator2

What it does

Inferelator2 is a inference algorithms that will take several different types of data, including time series, and create a dynamical model of the gene network to explain the time series. In general, the more data you can give to the algorithm the better it seems to perform (on simulated tasks).

Data

• Time series data (required)
• Steady state data (required)
• Knockout data
• Other data (knockdown, overexpression, multifactorial, etc)

Parameters

num_cores

The number of cores to use. The bootstraps are distributed amongst the cores, so it can greatly speed up the algorithm if doing a large number of bootstraps.

num_bootstraps

The number of bootstraps to run. Rule of thumb is that for a smallish dataset (50-200 genes), use around 50 boot straps. These get averaged at the end of the algorithm to form the final network.

delta_t_max and delta_t_min

Maximum time and minimum time gap to use for the time series. The time distance between the time steps in the time series must fall between these two parameters or that time point will be treated as steady state.

nCv

Number of crossvalidations to run. Default 10.

tau

A tradeoff measure on how much to take into account the amount of time between time points during inference. Default is 45, which seems to work well in most simulated cases.

prior_weight

The amount of weight that should be used with a prior. Set to 0 if not using a prior (default). With many bootstraps, it does not seem that a prior helps too much and the algorithm will generally converge on the same network regardless of the prior.

Example call

inf.setup(None, ss_data, settings, timeseries_data, multifactorial_data, "Inferelator2_TS_SS_NOPRIOR_NOPERT".format(meth, "inf_test"), None, prior=None, split_ts=False, leave_outs=leave_out)