The primary means of controlling the data modelling function is via data sources label for traits. However this is quite inflexible when creating multiple models that should run against different sub sets's of data since a trait can only belong to 1 data source.