I am not familiar with the "subject predicate object method" but I read your post with considerable interest (it is a topic that I am running into quite a bit lately) as I was struck by how much your treatment of directionality has an obvious parallel with verb treatments of subject and object.
To say "is a pet subject of" requires a subject and an object. In language, we depend either upon context (which can be quite ambiguous) or upon the rules of order to distinguish what is the pet subject, and who's pet subject it is. Because the brain plex is non-directional, both approaches break down.
### Pragmatic Approach ###
Use commas in child thoughts to add context to your plex. For example in PB, the child thought "Chuck Frey, pet projects", appears differently depending upon whether or not its parent thought: "Chuck Frey" is active. (Try this if you are unfamiliar with the method). So, I can associate many discrete subset aspects of Chuck Frey, each as an independent child thought of Chuck Frey. Then, I can create the thought "Leprechaun rehabilitation" with a thought type of "worthy causes", and link it to "Chuck Frey, pet projects". This contextually preserves the direction of the link.
I've gone a long way using this approach. With some restructuring, you find your major thoughts principally linked directly to sub-category thoughts that each collect the details relevant to that subcategory.
### Feature request approach ###
In a plex with very specific and narrow goals, one might create a thought "Chuck Frey" (thought type "subject"), linked (with a link type "pet project of") to thought "Mind Mapping" (thought type "object"). But what if you wish at the same time to represent "Mind Mapping benefits from directional semantic links". Then "Mind Mapping" is the subject and "directional semantic links" is the object. We need "Mind Mapping" to have two thought types to make this technique work. And as the number of links increases, it is reasonable to expect that the number of link types and thought types required would expand also. Is it then reasonable to conclude that thought types are most meaningful in relation to the thought's links?
In PB, thought types currently are assigned monolithically ( a thought has one and only one thought type, no matter what context the links to the thought seek to establish). In the real world and in our minds, items are rather more relative, so PB might benefit if thoughts could have an unlimited number of thought types as properties, and if the thought type displayed could be dependent upon the link type context.
What would that look like? Perhaps a thought might have a small cloud of thought/link type properties hovering about it, with each property in proximity to the link for which it is relevant.
Be vewy quiet. I'm hunting wabbit.