My 2 cents, for what it's worth. I'll just say that optimizations/performance are perhaps the biggest reasons why you'd use a graph database rather than APIs. Often the database will handle sharding and scaling and elasticity. A couple of graph databases on the market have Zeppelin/Jupyter notebook compatibility, allowing you to do some python stuff, then some SPARQL stuff, then some more python on the same triples. That might be a way to go.
