NETWOR KPROPERTIE SO FINTRINSICALL YMULTIVARIAT EPREDICTIV EGENES

A set of predictor genes is said to be intrinsically multivariate predictive (IMP) for a target gene if all properly containe dsubset so fth epredicto rse tar eba dpredictor sof the target but the full predictor set predicts the target with great accuracy. A basic measure of the predictive power of a set of genes with respect to the target is the Coefficient of Determination (CoD). In a previous paper, we described analytically the main properties of IMP genes introducing the IMP score, a metric based on the Coefficient of Determination (CoD) as a measure of predictiveness of the target gene. Basically, the IMP score depends on four main properties: logic of connection, predictive power, covariance between predictors, and marginal predictor probabilities (biases). This paper presents briefly a simulation study that seeks to capture correlations between IMP score and network properties. The results confirmed facts that were known previously from our analytic study, as well as new properties, such as the fact that gene “territory” size has positive correlation with IMP score.