Results on the DREAM 3 competition data.

Performance on the in silico data for challenge 4 of the DREAM 3 competition [8], [42], [43]. Team 315 was the top performer in the challenge [10], while teams 304 and 256 focused on time series and were among the top performers. Team 304 used Inferelator 1.0 as part of a larger reconstruction pipeline [19]. Team 256 also fit non-parametric ODEs though their approach differed from ours [29]; team 256 did not participate in the 100-node reconstructions due to the complexity of their method. The competition data includes observations for wild-type, knockouts of each gene, heterozygous knockdowns of each gene, and time series under multifactorial perturbations. There are 21 observations on each of the 4 time series for the 10-node networks and the 46 time series for the 100-node networks. Our reconstructions used knockout and time-series data only. PR and ROC values were computed using GeneNetWeaver.