National Institutes of Health (NIH)

Empower biomedical knowledge graphs with temporal reasoning!

Join the challenge team for live office hours to ask questions, clarify challenge requirements, and get guidance on the submission process.

It's About Time Office Hours Sept 24, 2026

Participation Agreement

Phase 1 aims to identify promising concepts and approaches for enabling temporal reasoning in KGs.

Phase 1 submissions are expected to include:

* Temporal Reasoning Frameworks/Designs: Conceptual frameworks or algorithmic designs for representing and reasoning over temporal aspects of biomedical knowledge
* Innovative Temporal Modeling Approaches: Clearly-articulated approaches for modeling temporal constructs such as events, sequences, durations, validity periods, and evolving evidence
* Use Case and Potential Impact: Identification of one or more relevant biomedical, clinical, or public health use cases that motivate the proposed approach
* Feasibility and Integration: Discussion of how the proposed approach could integrate with existing KG standards, ontologies, or infrastructures

Innovation in temporal representation of events, sequences, durations, validity intervals, evidence evolution, and uncertainty, including temporal embeddings. Soundness of temporal representation (event ordering, durations, validity intervals) and technical rigor of the temporal/dynamic-graph modeling. 

Temporal Modeling Innovation

Reasoning correctness, including temporal consistency. Inference uses only information available up to the query time (no temporal leakage) and respects event ordering and validity intervals.

Reasoning Correctness & Temporal Consistency 

Submission of ontology in a format that allows for logical inference (e.g., OWL, other description-logic or First Order Logic based language). Semantic modeling consistent with the Biolink Model and use existent ontologies (e.g., from the NLM or OBO); and existence of machine-verifiable axioms beyond the class hierarchy.

Ontology Rigor & Interoperability 

Feasibility of the approach to generate a semantically-rich integrated knowledge graph and the feasibility of integration with existing biomedical ontologies, standards and infrastructures where possible.

Feasibility & Integration 

Addresses a real-world problem in a specific clinical or public-health workflow. Execution of the solution can lead to meaningful alteration of decision making, Execution of the solution has the potential of producing measurable and positive impact on outcomes.

Clinical or Public-Health Workflow Relevance & Impact 

Phase 1: Conceptual Design and Feasibility

Phase 2 focuses on implementation, evaluation, and demonstration of the proposed approaches.

Only Phase 1 winners are eligible to submit to Phase 2.

Phase 2 submissions are expected to demonstrate:

* Implemented Temporal KG Methods: Working prototypes or tools that implement temporal representations, embeddings, and reasoning capabilities in a KG
* Demonstrated Temporal Reasoning: Evidence that the approach enables realistic or real-world scenario time-respecting inference, longitudinal reasoning, or evolving evidence interpretation
* Applied Use Cases: Demonstrations using realistic biomedical data or scenarios
* Evaluation and Analysis: Quantitative or qualitative evaluation showing improvements over static KG approaches (e.g., correctness of inference, temporal consistency, interpretability), including discussion of strengths, limitations, and assumptions
* Reusability and Extensibility: Documentation and artifacts showing that the approach can be adapted to other biomedical domains or datasets beyond the demonstrated use case

Demonstrated temporal reasoning on realistic biomedical data, including time-respecting and longitudinal inference; effectiveness is measured by performance on the shared benchmark tasks and test data, not by presentation quality.

Demonstrated Temporal Reasoning 

Correctness of inference verified on the test data, including temporal consistency and avoidance of temporally implausible inference (e.g., asserting a fact at a time before its supporting evidence holds, or violating a known validity interval).

Correctness of Inference 

A head-to-head comparison against a defined static-KG baseline, reported as qualitative or quantified improvement on the metrics listed below with clear documented methods, assumptions, and limitations. “Clear advantage over static KG approaches” means a material, quantified improvement over the baseline. Metrics required: Logical Inconsistencies/Unsatisfiable classes, Amount of Data (number of triples), Relationship Richness, Attribute Richness, Class/Property Instantiation Ratio, and Interlinking Completeness.

Evaluation Rigor 

Reusability of the approach for other biomedical domains or datasets beyond the demonstrated use case, evidenced by documentation and reusable artifacts.

Discover your next freelance opportunity

It’s About Time: Temporal Reasoning in Biomedical Knowledge Graphs Challenge