strange behavior in PPI data of string-db

Question

[aslan.mehrabi]

I have downloaded the homosapiens PPI network from the string-db.

I am using the total score which is assigned to each of the edges of this network. The strange observation from this data is lots of edges with the value of 900 which can be seen in the below histogram of edges of this network. (only the edges with weight more than 400 were considered)

Can any body please explain why this happened? Are these values acceptable?

[Jean-Karim Heriche]

Most of these probably come from the "prediction" methods like co-expression, neighborhood... Depending on what you want to do with the data, I would advise to select only the relevant interaction types.

Answer 1

[Lars Juhl Jensen]

No single piece of evidence is allowed to score greater than 900 in STRING. For this reason, any edge that is supported exclusively by one strong piece of evidence (e.g. a manually curated interaction from a pathway database) will have a score of 900.