Thus relevance parameters provide a principled way to decide which genes might be under the influence of a regulator: if the relevance parameter is small the gene response is unlikely to depend on the TFA of this regulator.
If the relevance parameter of one of these output GPs is low it means that the gene is unlikely to be regulated by the regulator.
The relevance score denotes the linear relevance parameter for gene g, i.e., β2, g, stands for the nonlinear relevance parameter β4, g (for β2, g and β4, g see equation (1)) and is the DosR-binding motif score as described above.
Consequently, we combine the relevance parameters of a gene with a score for the presence of the DosR binding motif in a logistic regression to improve on the prediction whether a gene is regulated by DosR or not.
After estimation of the TFA of DosR based on a number of known target genes using the GP model, we predict novel DosR-regulated genes: the parameters of the model are interpreted as relevance parameters indicating an existing functional relationship between TFA and gene expression.
By combining the relevance parameters derived from a GP model and sequence information, we confirmed all the documented DosR regulated genes.
β2, g and β4, g are referred to as relevance parameters in this paper.
The sparse priors amount to an automatic relevance determination (ARD) mechanism embedded in the GP model in that the estimated relevance parameters for genes not under DosR control are encouraged to be close to zero [ 22].
In the training phase, both the smoothness (relevance) parameters as well as the hidden values of the regulator need to be estimated.
As outlined in the introduction and detailed in the methods section, estimated relevance parameters of the output GP can be used to assess the strength of the influence of the reconstructed TFA of DosR on genes.
