text mining bug? Imatinib Treats Coronary Artery Disorder

[gglusman]

1) Search for what drugs may treat 'heart disease' 2) filter results for 'imatinib' - it has a very low score of 0.8, but nevertheless... 3) look at the 4th path displayed: imatinib treats coronary artery disorder subclass_of heart disease 4) click on 'treats' to get evidence 5) the first paper (of two) is https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7818412/ - a case report, "Successful kidney transplantation in a patient with pre‐existing chronic myeloid leukemia treated with imatinib." 6) search 'coronary' in the paper, finds just one mention: "Given the optimal response to imatinib treatment with a stable course of disease and the patient's good general state of health despite significant comorbidities such as arterial hypertension, insulin‐dependent diabetes mellitus, coronary artery disease, asymptomatic peripheral arterial disease, and cerebrovascular disease with a history of stroke, the patient was considered eligible for retransplantation"

In other words, CAD is mentioned as one comorbidity of that patient, and there's no indication that imatinib treats CAD. Moreover, imatinib is considered potentially cardiotoxic.

[andrewsu]

I tried to rerun the query https://ui.test.transltr.io/results?l=Coronary%20Artery%20Disorder&t=0&q=8a4ab001-3e8e-4690-92a3-539c9c1dc0bc but unfortunately imatinib no longer shows up in the results. However, it looks like this record from text mining provider is the one that came through previously https://biothings.ncats.io/text_mining_targeted_association/association/caac13f5f0d9dd679cf2e30e8304350ea0774981ec1775f74cd1503bca2e3853-treats. assigning to @bill-baumgartner...

[cbizon]

@gglusman please retest and close if appropriate

[bill-baumgartner]

My vote would be to leave open for now as TMKP has not addressed the issue yet.

[gglusman]

I retested and have a couple of novel observations.

When asking what drugs may treat Heart Disorder, imatinib is not among the answers. => https://ui.ci.transltr.io/results?l=Heart%20Disorder&i=MONDO:0005267&t=0&q=1ee0b9e4-f5ee-47e2-b446-af5bf94cbf8f

When asking what drugs may treat Coronary Artery Disorder (CAD), imatinib is listed among the 1132 results, with a score of 87.0 this time. Last time it had a score of 0.8 ! => https://ui.ci.transltr.io/results?l=Coronary%20Artery%20Disorder&i=MONDO:0005010&t=0&q=5032617c-fbd1-489a-9fe6-27d7b5bcfac4 There are now 7 paths connecting the drug to the disease. Focusing on the direct path, 5 references are cited in support: The last two are the same paper (Nozawa 2014), once with a PMID and once with a PMCID. Known issue. BUT, Nozawa 2014 does NOT support the assertion, either.

The new problem I see is that the other three publications don't talk about CAD at all... but they cite Nozawa 2014. The title of Nozawa 2014 in the References section is the only place where CAD is mentioned. Perhaps the References section should not be included when text-mining for the content of a paper? The paper cannot be construed to be claiming anything in its references.

[andrewsu]

This answer of imatinib - treats - CAD appears to be coming from ara-robokop, answer 66: https://arax.ci.transltr.io/?r=6d5242aa-6f76-446e-9ecd-4922d0a36844. It comes from the same underlying TMKP edge https://biothings.ncats.io/text_mining_targeted_association/association/caac13f5f0d9dd679cf2e30e8304350ea0774981ec1775f74cd1503bca2e3853-treats. So from the TMKP perspective, I think it is the same underlying issue as noted in the original post.

But what I find curious about this case is that the cited evidence in the Translator UI, the ARAX UI and the robokop doesn't seem to match what's returned by the TMKP API. That's illustrated in the screenshot below, which are snippets from these API calls:

• ROBOKOP response https://arax.ci.transltr.io/api/arax/v1.4/response/6d5242aa-6f76-446e-9ecd-4922d0a36844
• TMKP API call: https://biothings.ncats.io/text_mining_targeted_association/query?q=subject.id:%22DRUGBANK:DB00619%22%20AND%20object.id:%22MONDO:0005010%22:

I don't think the original TMKP pubs are any better at supporting the assertion that imatinib treats CAD, but I'm pointing this out just so the ROBOKOB team can make sure that the provenance is being handled as intended. (Also, I think ROBOKOP is using infores:textminingkp, which I think should be infores:text-mining-provider-targeted.)

[cbizon]

@EvanDietzMorris is this a known issue that you are working on?

[EvanDietzMorris]

Yes, we do have the wrong infores for tmkp edges in the robokop graph at the moment but it will be fixed in the next graph very soon.

On Thu, Jun 29, 2023 at 6:27 PM cbizon @.***> wrote:

@EvanDietzMorris https://github.com/EvanDietzMorris is this a known issue that you are working on?

— Reply to this email directly, view it on GitHub https://github.com/NCATSTranslator/Feedback/issues/334#issuecomment-1613989135, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE5TMUITYNQ57IBMWESWMF3XNYTP5ANCNFSM6AAAAAAY6CP4FA . You are receiving this because you were mentioned.Message ID: @.***>

[cbizon]

@EvanDietzMorris I think that this is a different issue. Take a look at Andrew's comment above. Is this because we maybe have an old pull of TMPK?

[cbizon]

At any rate, I don't think that this is an "opposite of what I asked for", so I'm removing the label

[gglusman]

Re-tested on CI August 12. The wrong answer is still there, with low score again.

The evidence for the direct path is just a link to BTE.

Evidence supporting 'imatinib treats heart failure' is one publication without displayed information: There is a link to a PMC paper though, and it does talk about imatinib... being cardiotoxic, as expected. Not treating heart failure. Other paths also include articles clearly stating that imatinib is cardiotoxic. This is the opposite of what I asked for.