Hereditary Hemorrhagic Telangiectasia OpenScientist Report Review

Date: 2026-04-24

Scope

Review of the verbatim OpenScientist report at research/Hereditary_Hemorrhagic_Telangiectasia-deep-research-openscientist.md against the current curated YAML, the Falcon report, and fetched primary references used during HHT curation.

Overall assessment

• The report is useful as a lead-generation artifact for disease-level structure.
• It is not safe for direct ingestion without review.
• The strongest value was in surfacing the 2024 PATH-HHT pomalidomide trial.
• The report also contains at least one ontology-anchor error and several claims that are plausible but too loosely sourced for direct promotion.

Findings That Held Up On Review

• Core disease framing as an autosomal dominant vascular dysplasia involving the BMP9/BMP10-ENG-ALK1-SMAD4 axis.
• Two-hit / focal lesion framing for AVM formation.
• Broad genotype-phenotype split:
• ENG enriched for pulmonary and cerebral AVMs.
• ACVRL1 enriched for hepatic AVMs.
• High disease-burden framing around recurrent epistaxis, iron deficiency, and visceral AVM complications.
• Therapeutic significance of the PATH-HHT randomized trial (PMID:39292928).

Review Findings

1. Incorrect ontology anchor in the verbatim report

The OpenScientist report uses MONDO:0008535 as the HHT identifier. The current validated repo entry uses MONDO:0019180 in kb/disorders/Hereditary_Hemorrhagic_Telangiectasia.yaml. This means the verbatim report should be treated as informative text, not as ontology-safe structured input.

2. One claim was clearly worth promotion after primary-source check

The report surfaced the pomalidomide Phase 3 result. After fetching the underlying abstract (PMID:39292928), that evidence was strong enough to add a new Pomalidomide Therapy treatment entry to the HHT YAML.

3. Several claims remain "interesting but not yet curation-safe"

These may be correct, but they were not promoted from the OpenScientist report without additional source-level checking:

• pregnancy-risk summary values
• manganese-deposition / basal-ganglia MRI discussion
• sex-difference claims
• forward-looking therapeutic language such as "potential first-ever FDA-approved therapy"

These are exactly the kinds of statements that should stay in the verbatim file until a reviewer fetches and checks the underlying papers.

4. Technical note on the provider run

For this HHT job, the OpenScientist /status endpoint continued to report running even after the artifacts ZIP already contained final_report.md. The verbatim report recovered cleanly from the ZIP, but this behavior means status alone is not a reliable completion signal for eval bookkeeping.

What Was Promoted From The Review

• PMID:39292928 was fetched and used to support Pomalidomide Therapy in the curated HHT YAML.

Specific Paper Check: PMC12274349

Paper:

• PMC12274349
• PMID:40681766
• DOI: 10.1038/s42003-025-08461-6
• Title: Overlapping upstream ORFs ending at c.125 lead to reduced Endoglin, contributing to Hereditary Hemorrhagic Telangiectasia.

Result:

• This paper was not surfaced in the verbatim OpenScientist HHT report.
• It was also not surfaced in the verbatim Falcon HHT report.
• It is not currently referenced in the HHT YAML.

Evidence for that conclusion:

• repo search across the four HHT report artifacts found no hit for PMC12274349
• no hit for PMID:40681766
• no hit for 10.1038/s42003-025-08461-6
• no hit for the title string

Interpretation:

• This omission is understandable. PMID:40681766 is a narrower 2025 molecular diagnostics / ENG 5'UTR mechanistic paper, whereas both agent reports leaned toward disease-level reviews, cohort studies, and clinically central management papers.
• It is still potentially useful for a future refinement of the ENG genetics section, especially if we want better coverage of noncoding pathogenic mechanisms and molecular diagnosis edge cases.

Bottom Line

Keep the OpenScientist file as the verbatim eval artifact. Use the review file to record what survived cross-checking, what was promoted into YAML, and what still needs source-level verification before curation.