Hints for the In Silico Challenge (DREAM4, Challenge 2)

Robert Prill and Daniel Marbach - IBM, EPFLAug 20, 2009 - 5:51 pm

A participant asked a few questions about the nature of the in silico data that we answered. Just so everyone is on equal footing, here is a transcript.

>> 1. Are the initial points for all of the time series
>> (*timeseries.tsv) the same steady state, up to noise?

In principle yes. The perturbation is applied at t=0, but did not yet
have an effect. Thus, the initial points of each time series are
independent samples of the wild-type steady state.

>> 2. Are the last time points (t=1000) the same as the initial state,
>> up to noise?

Not necessarily, because the time from t=500 when the perturbation is
removed until t=1000 may not be sufficient for the networks to
completely go back to the initial steady state. Also, in exceptional
cases the network could go to a different attractor after the
perturbation. We did not analyze whether this actually occurs, so this
possible issue should be considered part of the challenge.

>> 3. More generally, the description of the data leads one to assume
>> that both the wildtype and the single deletion networks have a
>> unique steady state. Is this assumption correct?

We generated the data using numerical simulations, without a detailed
analysis of the different attractors that the in-silico gene networks
have. Some networks probably have several steady states. The wild-type
steady state was defined arbitrarily as one of possibly several steady
states of the network. The steady states for the genetic perturbations
(knockouts and knockdowns) are those that the network converged to
from this wild-type steady state after applying the genetic
perturbation.

Note that the networks typically converge to a steady state, but as in
a biological experiment, there is no absolute guarantee. In
exceptional cases, there may be oscillations in the in silico
networks. Again, this possible issue is part of the challenge.

> 4. One more question: in the time series experiments, are there 5
> distinct perturbation effects (chemicals), or a single perturbation
> at 5 different concentration levels?

Each time series corresponds to a distinct perturbation.