Reactome physical association pairs in the human protein-protein interactions data

Question

[ebrudermanver]

I am using the data at the "Human protein-protein interactions" link on the Reactome web page. Here is the link to the actual data. My first question is: is this data representing directed interactions? And here is a second, a followup question. I realized there are cases where both A<physical association>B and B<physical association>A pairs exist in this data, where A and B are two proteins and physical association is the Interaction type. Does this mean both directions of interactions exist, i.e. A acts on B and B acts on A?

[Jean-Karim Heriche]

The interactions are automatically inferred from the pathways so it could explain why one sees A-B and B-A. Generally there's no reason to think of binary physical interactions as being directed.
It seems the Reactome documentation for the tab-delimited files is outdated. I think "physical association" is meant to replace the annotations "direct complex" and "indirect complex" which, as the docs note, were meaningless.
If you're interested in protein-protein interactions in general (I.e. not limited to curated pathways), I would suggest to use a more comprehensive resource such as iRefIndex or if it's too old compile your own from multiple sources (e.g. use the same as iRefIndex).