We address the problem of characterizing breast cancer, which today is done using staging guidelines. Our demo will show di erent breast cancer staging results that leverage the Whyis semantic nanopublication knowledge graph framework [8]. The system we developed is able to ingest breast cancer characterization guidelines in a semi-automated manner and then use our deductive inferencer to generate new information based on those guidelines as described in our ISWC resource track paper `Knowledge Integration for Disease Characterization: A Breast Cancer Example' [11]. In this paper we demonstrate the versatility of our framework using a synthetic patient pro le.
There are existing ontological representations for cancer characterization based
on previous AJCC cancer staging editions [
Some patients who do not respond as expected to traditional treatment paths may require the physician to consider additional testing measures, new sources, and/or alternative treatment plans. We will exemplify the interplay of disease characterization and personalized medicine being o ered by our cancer characterization tool, powered by semantic web tools, in our demonstration.
Let's consider the patient pro le with the biomarkers1 given in Table 1. Suppose there is a physician considering this multitude of parameters related to tumor biology as well as standard pathology to inform a diagnosis, treatment, and monitoring plan for this patient. With the utilization of the SDD methodology, the data in Table 1 is converted to the nanopublication format in the knowledge graph. Speci c to this use case, our cancer staging ontologies include the axioms in Listings 3.1 and 3.2 that were extracted from the AJCC 7th and 8th editions respectively. 1 The abbreviations used include: HER2 (Human Epidermal Growth factor receptor 2 ), ER (Estrogen Receptor), PR (Progesterone Receptor).
Once these axioms, along with the other 400+ staging axioms, are applied on the patient nanopublication data, the stages as per the 7th and 8th editions are determined (IIIA in 7th and IIB in 8th). Because of the additional data streams considered in the 8th guideline (i.e. Tumor Grade, HER2, ER and PR), the patient now has improved prognosis. The visualization tool in Fig. 1 shows the changes to the treatment and monitoring options based on the new stage.
More information on our system, including a video of the system demonstration, is available at https://cancer-staging-ontology.github.io. 4
With new understanding of cancer biology, guidelines are expected to increase in complexity and personalized care is more sought after as a component of treatment. Similarly, as the data streams for diagnosing and treating cancer patients becomes complicated, physicians may have to consult many di erent trusted sources and use knowledge from clinical trials and literature to decide on alternative treatment options which can take a great deal of the doctors' and the patients' precious time. Using our cancer characterization tool, physicians have access to patient records in RDF nanopublication format, and can infer the stage, as per the cancer staging ontology, that models the new staging guidelines, and then can investigate alternative, evidence-based treatment options. Our visualizations present evidence-based, updated staging determinations and treatment options, along with provenance. Physician-facing software applications can use our cancer characterization tool to provide physicians with an e cient way to investigate alternative treatment options based on di erent staging guidelines.
In the future, new guidelines for cancer staging are expected to incorporate genomic test results analyzed in the context of the patient's history. Many predict a rapid in ux of information related to cancer from clinical trials, as well as from basic science research. Leveraging all these heterogeneous data sources and making connections to understand the data is of utmost importance. Our system demonstration shows how semantics are being used to support this fast changing landscape.
This work is partially supported by IBM Research AI through the AI Horizons Network. We thank our colleagues from IBM (Amar Das, Ching-Hua Chen) and RPI (John Erickson, Alexander New, Rebecca Cowan) who provided insight and expertise that greatly assisted the research.