A common practice for query performance optimization in relational databases systems is materialized views.
A view is a subset of your data that (usually) cannot be updated directly but reflects the superset of the data it is based on. A materialized view is an extension to this concept by not only generating a view on a subset of data, but also storing the representation of that view to persistent storage. This especially makes sense if the construction of such a materialized view is computationally expensive. This is because, in subsequent queries making use of such a view, the amount of time that the creation of such a materialized view would have taken can be saved in each of those queries.
The same concept also applies for graph databases. Common subgraphs, which are used among multiple queries, can be defined as materialized views. Therefore, graph queries making use of these subgraphs can benefit using them.
In GraphX, only subgraphs with a triplet structure can be used but GraphFrames supports arbitrary subgraph structures. Materialized views in GraphFrames can be defined manually and all queries automatically make use of them.