Even though my data is undirectional graph can the explainer still give directed graph exaplantion

[bojack-horseman91]

I have checked that all my data is undirectional using is_directed() function and found none are directional and yet when I use GNNExxplainer to produce an explanation I get a directed graph visualized when I run explanation.get_explanation_subgraph().visualize_graph("path.png"). Can you tell me what's wrong?
P.S: I have used PNAConv as my GNN then feed it to a Fully connected layer

[rusty1s]

We visualize a directed graph since the scores for an edge may still be different depending on its direction. Imagine you have a one-layer GNN, then you would expect that all outgoing edges from a seed node have importance score ~ 0 (since they do not contribute to the final output), while incoming edges have an importance score > 0.

[bojack-horseman91]

So for interpretation purpose how do we percieve these directions? Because my edges are representing relationship between two genes how do i make inference from the direction of the graph ?

…

On Thu, 2 May, 2024, 6:55 pm Matthias Fey, ***@***.***> wrote: We visualize a directed graph since the scores for an edge may still be different depending on its direction. Imagine you have a one-layer GNN, then you would expect that all outgoing edges from a seed node have importance score ~ 0 (since they do not contribute to the final output), while incoming edges have an importance score > 0. — Reply to this email directly, view it on GitHub <#9261 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ARAQLCJTPEQ5MZPVSFM4ZXTZAIZVFAVCNFSM6AAAAABHAEWTQ6VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TEOJVGMZDQ> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[rusty1s]

I would just interpret them as information flow. One edge updates node A by information from node B, and one edge updates node B by information from node A.

how do i make inference from the direction of the graph ?

Can you clarify? I don't understand yet.

[bojack-horseman91]

Well I was hoping to get some biological inference because I had set the edge attribute to be co-methylation ,which is a fancy way of saying correlation between two genes. How do i interpet the direction in a biological terms?

…

On Fri, 3 May, 2024 at 6:46 PM, Matthias Fey ***@***.***> wrote: I would just interpret them as information flow. One edge updates node A by information from node B, and one edge updates node B by information from node A. how do i make inference from the direction of the graph ? Can you clarify? I don't understand yet. — Reply to this email directly, view it on GitHub <#9261 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ARAQLCJ2ABGM7ZRZUQ5GPRLZAOBJDAVCNFSM6AAAAABHAEWTQ6VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TGMBVGM3TM> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>