how to interpret the trees?

[SeongwooLimKR]

When I run the BoostSRL without "-mlnClause" option, I got the following tree with Drug-Drug Interaction (DDI) dataset.
The location of tree file is "$MY_RESULT_DIR/bRDNs/Trees/InteractsTree0.tree"

My questions are
(1) How to interpret "_"?
Can I understand the first rule

(Interacts(A, _, 0.29566210336691023) :- /* #neg=46 #pos=110 */ TransporterInducer(A, _), TargetInhibitor(A, _), !).

as

(Interacts(A, B, 0.29566210336691023) :- /* #neg=46 #pos=110 */ TransporterInducer(A, B), TargetInhibitor(A, B), !).

or

(Interacts(A, B, 0.29566210336691023) :- /* #neg=46 #pos=110 */ TransporterInducer(A, C), TargetInhibitor(A, D), !).

(2) How to interpret "UniqueVar"?
Is the third rule

(Interacts(A, B, 0.4228365702020148) :- /* #neg=110 #pos=190 */ EnzymeInducer(A, UniqueVar1), EnzymeSubstrate(A, UniqueVar1), Enzyme(UniqueVar1, B), !).

same with

(Interacts(A, B, 0.4228365702020148) :- /* #neg=110 #pos=190 */ EnzymeInducer(A, C), EnzymeSubstrate(A, C), Enzyme(C, B), !).

(3) Is there any code or tool to convert from the rules generated by BoostSRL to the rules with Alchemy or Tuffy format?

Thank you,
-Seongwoo

[harshakokel]

Hello Seongwoo,

As always, thank you for your interest in our work. Please find the answers below.

Q1: Underscores are interpreted as anonymous variables (like prolog). Their exact grounding would not matter. Among the two interpretations you mention, it would be closer to the second one

(Interacts(A, B, 0.29566210336691023) :- /* #neg=46 #pos=110 */ TransporterInducer(A, C), TargetInhibitor(A, D), !).

Q2: Your interpretation of UniqueVar is correct. That is, all occurrence of UniqueVar1 are same variables.

Q3: BoostSRL generates a folder called WILLtheories/. The rules printed there are closer to the Tuffy format. I would recommend looking at that.

Note: BoostSRL run without '-mlnClause' flag, is essentially learning an RDN model.

Hope this helps.
Regards,
Harsha