Fig. 7

To demonstrate the use of MOFGalaxyNet and GCN node classification for predicting guest accessibility, consider an example. The edges of the graph are depicted with different colors and thicknesses, which correspond to the connection weights. Suppose we want to predict the PLD size for the AZADUC node. To do this, we obtain the feature values of its four neighbors, including AZADUC itself, and apply an aggregation function by taking their average. The resulting value is then fed into the GCN, which returns a PLD size as the predicted label for AZADUC