Subject of the Challenge:

Knowledge graphs (KGs) are foundational tools for biomedical research, clinical decision support, public health analytics, and AI-driven discovery. However, most existing biomedical and clinical KGs represent knowledge embedded as static factual information that lack the ability to encode how entities, relationships, and evidence may evolve over time. This limitation fundamentally hinders the accuracy, reliability, and trustworthiness of downstream analytics and AI systems that rely on knowledge in these graphs that may change over time.

In real‑world settings, nearly all biomedical and clinical processes are temporal. Care pathways unfold over months to years as patients move through diagnostic, treatment, and recovery. Diagnostic criteria and clinical guidelines change as new evidence emerges. Treatments are often sequential and not interchangeable, and their appropriateness depends on disease stage, prior response, comorbidities, and evolving standards of care. Scientific claims may be supported, refined, or contradicted as new data accumulate. Public health risks shift across waves of outbreaks, variants, and available interventions. Without explicit temporal representation, KGs cannot distinguish “what is true now” from “what was once true,” nor can they reason over longitudinal trajectories or time‑respecting causal paths to make future projections.

The absence of temporal embeddings and temporal reasoning mechanisms may lead to critical failures. Static KGs may conflate early‑ and late‑stage therapies, allow temporally impossible inference paths, obscure exposure windows in contact tracing, merge outdated guidance with current recommendations, or treat contradictory scientific claims as equally valid. These shortcomings can result in incorrect reasoning, misinterpretation of evidence, and clinically implausible or unsafe recommendations. As AI agents increasingly query KGs to support trustworthy decision‑making, these risks are amplified. Despite growing recognition of this challenge, most KGs and embedding methods still rely on static representations that are not designed to capture time. There is no widely adopted, interoperable approach for modeling temporal relationships (e.g., event sequencing, duration, validity intervals, evidence evolution, and uncertainty) in KGs.

This Challenge seeks to address this gap by incentivizing the development of innovative methods, tools, or frameworks that enable robust temporal embeddings and temporal reasoning in KGs. Proposed solutions should allow downstream analytics and AI systems to accurately interpret data in temporal context, reason over sequences of events, and distinguish evolving evidence, guidelines, and interventions. Solutions should be applicable to real‑world biomedical and clinical use cases and should align with and leverage existing standards where possible.

By advancing temporal KG capabilities, this Challenge aims to facilitate the development of new and novel methods, which unlock more accurate, context‑aware, and trustworthy AI‑driven insights that reflect how knowledge, evidence, and clinical workflows evolve over time.

This Challenge has two phases: 1) Concept Design and Feasibility, and 2) Prototype Implementation and Demonstration.

Statutory Authority to Conduct the Challenge

The Office of Data Science Strategy (ODSS) within the Office of the Director of the National Institutes of Health (NIH) is conducting this Challenge under the America Creating Opportunities to Meaningfully Promote Excellence in Technology, Education, and Science (COMPETES) Reauthorization Act of 2010, as amended [15 U.S.C. § 3719].

This Challenge is consistent with NIH’s authority to use prize competitions to stimulate innovation that has the potential to advance the agency’s mission, including by soliciting transformative solutions to important biomedical research problems. The Challenge directly supports the mission of the ODSS, which provides NIH-wide leadership for a modernized biomedical data ecosystem and catalyzes new capabilities in biomedical and health data science through strategic partnerships, advanced technologies, computational methods, data standards, interoperability, and reproducibility. Specifically, the Challenge seeks innovative methods, tools, and frameworks that enable temporal embeddings and temporal reasoning in biomedical knowledge graphs so that downstream analytics and AI systems can interpret data in temporal context, reason over sequences of events, distinguish evolving evidence and guidelines, and generate more accurate, context-aware, and trustworthy insights. These goals are closely aligned with ODSS priorities to advance data science infrastructure, emerging AI methods, interoperable data models, and reproducible computational workflows.