Inferring negative regulation by complex formation (sequestration)

[goodb]

Right now, the rule looks like this:
* If reaction2 is enabled_by entity1
* And reaction1 has entity1 as an input
* And reaction1 has entity2 as an input
* And reaction1 has complex1 containing entity1 and entity2 as output
* And NOT(reaction3 enabled_by | has_input complex1)
* Then infer that reaction1 directly_negatively_regulates reaction2
(by binding up the entity that enables reaction2 to happen).

The rule fires a lot (200+ assertions) for https://reactome.org/PathwayBrowser/#/R-HSA-3899300 . Would like input if this behavior is desirable. Here is an example of what is going on:

[deustp01]

And NOT(reaction3 enabled_by | has_input complex1)

NOT(reaction3... or NOT(reaction2... ?

[goodb]

Sorry, my parser isn't working properly here. Could you clarify?

My intent with the 'NOT' was to say that the complex generated by reaction1 doesn't go off to do something useful to another reaction3. So in the top reaction there, the rule should not activate if the output complex SUMO1:NCOR2 went on to be the input or enabler for another reaction.

[deustp01]

I'm not sure. I think the regulates relationship needs to have a single (regulated) target activity. Can't point to a case immediately, but we certainly have to allow for the possibility that a complex of proteins A and B positively regulates one target activity and negatively regulates a second target activity, and it looks like your logic only allows formation of the A and B complex to have only one kind of regulatory effect.

[goodb]

The formation of the complex could have other regulatory effects that would be captured by other rules.

I see your point though. Why should we stop this rule from activating if the complex has additional functions (as I imagine in reality almost everything probably does)? We should probably get some input from @huaiyumi here as he designed the rule last summer.

When it was first implemented, the example we used to test was inferring a negatively_regulates relation from the reaction "The ligand trap binds the ligand BMP2, blocking BMP signalling" to "BMP2 binds to the receptor complex". I see Reactome took out the enabler for BMP2 binds to the receptor complex and added the GO term "sequestering of BMP from receptor via BMP binding" to the ligand trap reaction. As a result the new GO-CAM for that one does not have the causal relationship between the corresponding function nodes. Are you happy with that result ?

Are we maybe going too far from Reactome's intent with this inference ?

[goodb]

Looking at more examples I think this rule definitely needs some reconsideration. e.g. for the pathway 'NTRK2 activates RAC1' it seems to be in conflict with another rule that suggest positive regulation. Just looking at the pathway (and its name) it appears like a positive regulation case, not a negative one.

The rule currently activates in 87 pathways:

R-HSA-8849474
R-HSA-196741
R-HSA-525793
R-HSA-4570464
R-HSA-3214858
R-HSA-5205685
R-HSA-171007
R-HSA-72163
R-HSA-1295596
R-HSA-3108214
R-HSA-182971
R-HSA-8963678
R-HSA-1221632
R-HSA-8875555
R-HSA-3371453
R-HSA-1433557
R-HSA-4615885
R-HSA-9032759
R-HSA-3232118
R-HSA-198753
R-HSA-189451
R-HSA-2162123
R-HSA-427413
R-HSA-416476
R-HSA-427359
R-HSA-5221030
R-HSA-5250924
R-HSA-399955
R-HSA-1660662
R-HSA-5653890
R-HSA-1663150
R-HSA-416550
R-HSA-977225
R-HSA-200425
R-HSA-418457
R-HSA-375165
R-HSA-2299718
R-HSA-75205
R-HSA-5627123
R-HSA-2467813
R-HSA-5675221
R-HSA-5625740
R-HSA-210993
R-HSA-8866904
R-HSA-140834
R-HSA-5625900
R-HSA-4551638
R-HSA-140875
R-HSA-140837
R-HSA-9033500
R-HSA-5627117
R-HSA-176974
R-HSA-1268020
R-HSA-380972
R-HSA-75035
R-HSA-3899300
R-HSA-5689603
R-HSA-4641262
R-HSA-5607764
R-HSA-844456
R-HSA-933542
R-HSA-5668541
R-HSA-2672351
R-HSA-982772
R-HSA-5689880
R-HSA-1059683
R-HSA-4755510
R-HSA-5357905
R-HSA-9020702
R-HSA-9010642
R-HSA-1169408
R-HSA-4655427
R-HSA-9027276
R-HSA-3928662
R-HSA-428890
R-HSA-428543
R-HSA-8849471
R-HSA-977443
R-HSA-5689896
R-HSA-4090294
R-HSA-4085377
R-HSA-392517
R-HSA-163359
R-HSA-5628897
R-HSA-202433
R-HSA-983695
R-HSA-6809371

[goodb]

The rule can be improved to avoid cases like the 'NTRK2 activates RAC1' example above by filtering out matches in cases where the target enabler is an active unit of the same complex that is the output of the first reaction. This reduces the 87 pathways to 56 and none of the 56 exhibit the property of having both negative and positive regulation happening between the same 2 reactions.

[goodb]

There remain many open questions on the 56 pathways that have matches to this rule. Unwinding of DNA is another example where its not clear to me that the rule is identifying regulation.

[goodb]

Here is an example where it looks like it is working. 'Regulation of TNFR1 signaling'

[goodb]

@deustp01 my impression here is that this rule may be of use to Reactome curators but is probably not consistently correct enough to be used in the automated conversion. Perhaps we can talk about this one on the call next week. (Unless something else came up in Cambridge that we should discuss.)

[goodb]

@deustp01 @huaiyumi @ukemi here is a list of the 56 pathways that are impacted by this rule as it stands: You can go direct to the go-cam for each by prepending the identifier with the noctua prefix e.g. http://noctua-dev.berkeleybop.org/editor/graph/gomodel:R-HSA-189451

hsa_id label n_generated_assertions
R-HSA-189451 Heme biosynthesis 1
R-HSA-1433557 Signaling by SCF-KIT 1
R-HSA-3928662 EPHB-mediated forward signaling 1
R-HSA-5627123 RHO GTPases activate PAKs 1
R-HSA-200425 Import of palmitoyl-CoA into the mitochondrial matrix 1
R-HSA-2299718 Condensation of Prophase Chromosomes 1
R-HSA-5627117 RHO GTPases Activate ROCKs 1
R-HSA-75035 Chk1/Chk2(Cds1) mediated inactivation of Cyclin B:Cdk1 complex 1
R-HSA-399955 SEMA3A-Plexin repulsion signaling by inhibiting Integrin adhesion 1
R-HSA-9027276 Erythropoietin activates Phosphoinositide-3-kinase (PI3K) 1
R-HSA-1660662 Glycosphingolipid metabolism 1
R-HSA-844456 The NLRP3 inflammasome 1
R-HSA-1059683 Interleukin-6 signaling 1
R-HSA-416550 Sema4D mediated inhibition of cell attachment and migration 1
R-HSA-5625740 RHO GTPases activate PKNs 1
R-HSA-977443 GABA receptor activation 1
R-HSA-5689896 Ovarian tumor domain proteases 1
R-HSA-5628897 TP53 Regulates Metabolic Genes 1
R-HSA-2467813 Separation of Sister Chromatids 1
R-HSA-5668541 TNFR2 non-canonical NF-kB pathway 1
R-HSA-1663150 The activation of arylsulfatases 1
R-HSA-392517 Rap1 signalling 1
R-HSA-525793 Myogenesis 2
R-HSA-4755510 SUMOylation of immune response proteins 2
R-HSA-5221030 Oxidative demethylation of DNA 2
R-HSA-4570464 SUMOylation of RNA binding proteins 2
R-HSA-198753 ERK/MAPK targets 2
R-HSA-8963678 Intestinal lipid absorption 2
R-HSA-1268020 Mitochondrial protein import 2
R-HSA-5625900 RHO GTPases activate CIT 2
R-HSA-428890 Role of ABL in ROBO-SLIT signaling 2
R-HSA-418457 cGMP effects 2
R-HSA-196741 Cobalamin (Cbl, vitamin B12) transport and metabolism 3
R-HSA-140875 Common Pathway of Fibrin Clot Formation 3
R-HSA-5357905 Regulation of TNFR1 signaling 3
R-HSA-176974 Unwinding of DNA 3
R-HSA-427359 SIRT1 negatively regulates rRNA expression 3
R-HSA-182971 EGFR downregulation 4
R-HSA-416476 G alpha (q) signalling events 4
R-HSA-4655427 SUMOylation of DNA methylation proteins 4
R-HSA-6809371 Formation of the cornified envelope 4
R-HSA-2672351 Stimuli-sensing channels 5
R-HSA-140837 Intrinsic Pathway of Fibrin Clot Formation 5
R-HSA-75205 Dissolution of Fibrin Clot 5
R-HSA-8866904 Negative regulation of activity of TFAP2 (AP-2) family transcription factors 6
R-HSA-140834 Extrinsic Pathway of Fibrin Clot Formation 6
R-HSA-2162123 Synthesis of Prostaglandins (PG) and Thromboxanes (TX) 8
R-HSA-4085377 SUMOylation of SUMOylation proteins 9
R-HSA-9033500 TYSND1 cleaves peroxisomal proteins 12
R-HSA-3232118 SUMOylation of transcription factors 21
R-HSA-4615885 SUMOylation of DNA replication proteins 24
R-HSA-1169408 ISG15 antiviral mechanism 30
R-HSA-4090294 SUMOylation of intracellular receptors 31
R-HSA-4551638 SUMOylation of chromatin organization proteins 37
R-HSA-3108214 SUMOylation of DNA damage response and repair proteins 76
R-HSA-3899300 SUMOylation of transcription cofactors 288

[goodb]

Note to check the comment on the asserted relation to ensure you are looking at one generated by this rule and not another regulation rule. The explanation in the comments for this rule will look like this: "The relation 'ALAD condenses 2 dALAs to form PBG' 'directly negatively regulates' 'ALAD binds to Pb2+' was inferred because: 'ALAD binds to Pb2+' has inputs A and B, 'ALAD binds to Pb2+' has output A/B complex, and 'ALAD condenses 2 dALAs to form PBG' is enabled by B. "

[huaiyumi]

I took a look at the 'NTRK2 activates RAC1' pathway in Reactome. It seems that the resulting complex is an enabler of the subsequent reaction. The issue is the active unit of the subsequent MF is not the entire complex, but a subunit of it. I wonder if that caused the rule to fail.

[huaiyumi]

[huaiyumi]

Reading through the previous comments, it think you guys already reached the same conclusion. I checked a few pathways in the list. It seems that all of them may due to this problem.

[goodb]

@deustp01 I checked again and the 2nd list mentioned in the comment above with the 56 pathways is ready for your perusal. (The issue with the active units was indeed already resolved, as you can see in the current noctua-dev model for NTRK2 activates RAC1.)

I suspect my next action here should be to take the rule out of the general purpose conversion and thus to close this ticket. We may consider moving this or something like it over into this project: https://github.com/geneontology/pathways2GO/projects/2 - which should get started once we judge the current goal of a general purpose, valid conversion of Reactome into GO-CAMs to be complete.

[goodb]

Conclusion here is that the rule is often, but not always correct. This provides useful feedback for the Reactome curators, but does not belong in this version of an automated translation. Moving issue to Quality Assurance and Feedback project https://github.com/geneontology/pathways2GO/projects/2

(note also to investigate odd inferences for R-HSA-8866904 )