Fig 1. Filtering algorithm diagram.

The filtering algorithm is performed in three phases. The first phase identifies features within the data types that have weak similarity with the outcome, the second phase identifies features that are weakly similar between the data types by summing the absolute value of similarities, and the third phase performs the final filtering step by removing any features that were identified as weak from phase 1 and phase 2. If features are weakly related to the outcome and weakly related with any features in the corresponding data type, then it is unlikely to be involved in the supervised network, and thus provides the motivation for SuMO-Fil. The phases are described in full detail in Algorithm 1.