The adaptive network weights are determined using a regression-like procedure and roughly represent the added predictive value (when considering all other selected networks) of the links in the network for predicting co-membership in the query list (for ‘Assigned based on query genes’) or co-annotation in Gene Ontology (for ‘Gene Ontology-based weighting’). In general, these network weights roughly represent the usefulness of the network for the prediction task; however, anecdotally, network weights have a small bias toward those that improve the precision of predictions rather than recall. As such, small networks with unique links can sometimes get surprisingly large weights.
