From 5da4bd9784a30f1eb6fb004dcd2798da67d884bb Mon Sep 17 00:00:00 2001 From: David Glymph Date: Mon, 22 May 2023 15:24:56 -0400 Subject: [PATCH] update documentation with info about bubble graphs --- src/pages/About.jsx | 2 +- src/pages/Guide.jsx | 2 +- src/pages/Tutorial.jsx | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/src/pages/About.jsx b/src/pages/About.jsx index 176e391a..c039dd9c 100644 --- a/src/pages/About.jsx +++ b/src/pages/About.jsx @@ -19,7 +19,7 @@ export default function About() {

Questions or queries are represented in ROBOKOP as JSON templates that have been abstracted into a more friendly UI. Each node in the query graph denotes a biomedical entity with a defined category (e.g., disease) and defined properties (e.g., breast cancer versus cancer); likewise, each edge denotes a predicate that can be specified to limit the allowable relationships between nodes (e.g., associated_with). Nodes and edges can be specified by way of text description, using the autocomplete drop-down menu, or, for nodes, by directly entering a CURIE (Compact Uniform Resource Identifier) (e.g., MONDO:0004989 for breast cancer).

The node categories and edge categories are defined by Biolink Model, which is an open-source data model and upper-level ontology that formalizes the relationships between biomedical entities such as gene, disease, chemical, and phenotype as a set of hierarchical interconnected categories and relationships between them or predicates, e.g., “chemical entity X causes disease Y” or “drug X treats disease Y”. Biolink serves as the “semantic glue” for the ROBOKOP application by enabling integration and harmonization across ROBOKOP KG’s diverse underlying knowledge sources. For more information on Biolink Model, please refer to Unni et al. 2022 and the Biolink GitHub website.

-

Query results are in the form of knowledge subgraphs or answers that match the categories and desired properties of the nodes and edges.

+

Query results are in the form of knowledge subgraphs or answers that match the categories and desired properties of the nodes and edges. In the Knowledge Graph Bubble, the size of each ‘bubble’ represents how relatively common that entity is represented among the full answer set.

Answer Scoring and Ranking

Questions or queries that include very few nodes and edges or that include many specified nodes and edges typically result in numerous knowledge subgraphs or answers. As such, the scoring and ranking of answers by relevance is critical for user analysis. ROBOKOP scores and ranks each answer within an answer set using a complex scoring algorithm. In brief, the ROBOKOP answer scoring-and-ranking algorithm weights each edge within each knowledge subgraph based on the number of supporting PubMed publications. The publication support is provided by either the curated knowledge source from which a particular edge was derived or by an additional ROBOKOP service, termed OmniCorp, which contains a graph of PubMed identifiers linked to node categories or biomedical entities co-occurring within PubMed abstracts. The ROBOKOP answer scoring-and-ranking algorithm treats publications derived from curated knowledge sources with greater importance than those derived from OmniCorp.

diff --git a/src/pages/Guide.jsx b/src/pages/Guide.jsx index 46e9e337..070b7897 100644 --- a/src/pages/Guide.jsx +++ b/src/pages/Guide.jsx @@ -59,7 +59,7 @@ export default function Guide() { Receive an Answer

- The ROBOKOP KG drives the ROBOKOP application and contains integrated and harmonized knowledge derived from dozens of public data sources. The ROBOKOP KG can be queried through the ROBOKOP UI or by direct Cypher query. Most users will find that the ROBOKOP UI provides a more convenient query tool than Cypher query and also allows users to more readily explore knowledge subgraphs or answers and associated provenance and publication support. When a user poses a question to the ROBOKOP application, ROBOKOP creates an ‘answer set’, which consists of a ranked list of potential answers to the question or query, derived using the ROBOKOP reasoning engine. + The ROBOKOP KG drives the ROBOKOP application and contains integrated and harmonized knowledge derived from dozens of public data sources. The ROBOKOP KG can be queried through the ROBOKOP UI or by direct Cypher query. Most users will find that the ROBOKOP UI provides a more convenient query tool than Cypher query and also allows users to more readily explore knowledge subgraphs or answers and associated provenance and publication support. When a user poses a question to the ROBOKOP application, ROBOKOP creates an ‘answer set’, which consists of a ranked list of potential answers to the question or query, derived using the ROBOKOP reasoning engine. Note that the size of each ‘bubble’ in the Knowledge Graph Bubble reflects how relatively common that entity is represented among the full answer set.

Note that the ROBOKOP KG is continuously evolving; as such, answers derived today may not be the same as answers derived tomorrow. This behavior is expected, as additional knowledge sources are integrated into the ROBOKOP KG and the reasoning engine matures. diff --git a/src/pages/Tutorial.jsx b/src/pages/Tutorial.jsx index 44417214..3058bae7 100644 --- a/src/pages/Tutorial.jsx +++ b/src/pages/Tutorial.jsx @@ -58,7 +58,7 @@ export default function Tutorial() { One-hop query results for all diseases associated with 2,3,7,8-tetrachlorodibenzo-P-dioxin in ROBOKOP.

- Users can then explore answers or knowledge subgraphs. In this example, the top-ranked answer is for neoplasm, with a score of 0.814. Clicking the answer path will display the answer knowledge subgraph in the Answer Explorer. + Users can then explore answers or knowledge subgraphs. Note that the size of each ‘bubble’ in the Knowledge Graph Bubble reflects how relatively common that entity is represented among the full answer set. In this example, the top-ranked answer is for neoplasm, with a score of 0.814. Clicking the answer path will display the answer knowledge subgraph in the Answer Explorer.

Exploring one-hop query results for all diseases associated with 2,3,7,8-tetrachlorodibenzo-P-dioxin in ROBOKOP.