Skip to main content

PubMedPipeline

What it is​

  • A pipeline that queries PubMed within a date range and returns results as an RDF Graph.
  • Can filter results to include only papers that have a downloadUrl (PubMed Central downloadable).

Public API​

Classes​

  • PubMedPipelineConfiguration(PipelineConfiguration)

    • Pipeline configuration type (currently no additional fields).

  • PubMedPipelineParameters(PipelineParameters)

    • Input parameters for running the pipeline.
    • Fields:
      • query: str — PubMed search query.
      • start_date: str — start date (string format expected by integration; CLI suggests YYYY-MM-DD or YYYY/MM/DD).
      • end_date: Optional[str] = None — end date; if None, searches up to present.
      • sort: Optional[Literal["pub_date","Author","JournalName","relevance"]] = "pub_date" — sorting mode.
      • downloadable_only: Optional[bool] = False — include only results with downloadUrl.
      • max_results: Optional[int] = 100 — max results (1..10,000).

  • PubMedPipeline(Pipeline)

    • Main pipeline class.
    • Methods:
      • __init__(configuration: PubMedPipelineConfiguration)
        • Initializes the pipeline and an internal PubMedIntegration(PubMedAPIConfiguration()).
      • run(parameters: PipelineParameters) -> Graph
        • Executes the PubMed query and returns an RDF Graph aggregated from result rdf() outputs.
        • Raises ValueError if parameters is not PubMedPipelineParameters.
      • as_api(...) -> None
        • Declared but not implemented (pass).
      • as_tools() -> List[BaseTool]
        • Returns a LangChain StructuredTool named search_downloadable_pubmed_papers that runs the pipeline and returns Turtle serialization.

Configuration/Dependencies​

  • Depends on:
    • naas_abi_core.pipeline: Pipeline, PipelineConfiguration, PipelineParameters, Graph
    • naas_abi_marketplace.applications.pubmed.integrations.PubMedAPI:
      • PubMedIntegration, PubMedAPIConfiguration, PubMedPaperSummary
    • LangChain: langchain_core.tools.StructuredTool
    • FastAPI: fastapi.APIRouter (only referenced by as_api, which is not implemented)
    • CLI dependencies when run as a script: click, rich

Usage​

Run from Python​

from naas_abi_marketplace.applications.pubmed.pipelines.PubMedPipeline import (
    PubMedPipeline,
    PubMedPipelineConfiguration,
    PubMedPipelineParameters,
)

pipeline = PubMedPipeline(PubMedPipelineConfiguration())
graph = pipeline.run(
    PubMedPipelineParameters(
        query="cancer biomarkers",
        start_date="2024-01-01",
        end_date="2024-03-01",
        downloadable_only=True,
        max_results=50,
    )
)

ttl = graph.serialize(format="turtle")
print(ttl[:500])

Use as a LangChain tool​

pipeline = PubMedPipeline(PubMedPipelineConfiguration())
tool = pipeline.as_tools()[0]

turtle = tool.run({
    "query": "machine learning radiology",
    "start_date": "2024-01-01",
    "end_date": None,
    "downloadable_only": False,
    "max_results": 10,
})
print(turtle[:300])

Run as a script (writes pubmed_output.ttl)​

python PubMedPipeline.py --query "diabetes" --start-date 2024-01-01 --end-date 2024-02-01

Caveats​

  • PubMedPipeline.run() only accepts PubMedPipelineParameters; passing any other PipelineParameters subtype raises ValueError.
  • as_api() is not implemented.
  • downloadable_only=True filters results by checking result.downloadUrl is not None before adding result.rdf() to the output graph.