Invention Reference Number
This invention provides a platform that uses AI agents and biological networks to uncover and interpret disease-relevant biological mechanisms. It helps researchers and clinicians identify which genes are involved in a disease, which drugs may help, and which model organisms can be used to test them. Unlike conventional systems, it works across multiple data types and species and explains its predictions in natural language summaries. It can be used in drug development, rare disease discovery, toxicity prediction, and personalized medicine.
Description
The central problem this invention addresses is the inability to reason across heterogeneous biological data to uncover conserved, testable mechanisms of disease, drug action, or toxicity. Traditional systems in translational genomics are gene-centric, siloed by data type, and limited in interpretability. They lack the capacity to integrate large-scale omics data with biological networks and evolutionary logic in a way that supports therapeutic decision-making, rare disease inference, and model system selection. This limitation hinders progress in areas like drug repurposing, safety pharmacology, regulatory biomarker development, and mechanistic diagnostics, especially when patient or disease data are sparse or noisy.
The invention integrates multiplex biological networks with random walk–based graph embeddings to generate "mechanistic clades"—topologically coherent gene modules reflecting functional biological processes. Large language model (LLM) agents, coordinated by a reinforcement-trained orchestration engine, interpret these clades in narrative, experimental, and therapeutic contexts. Cross-species conservation scoring identifies model systems in which these mechanisms are preserved, enabling translational planning. The platform also supports predictive inference, such as identifying side effect propagation pathways or repurposing opportunities, using agent outputs and reward shaping optimized for biological plausibility and graph coherence. The system is modular, interpretable, and adaptable to a wide range of disease contexts and data types.
Benefits
- Explainable outputs: Delivers clear, explainable outputs and ranked drug candidates, offering transparency compared to black-box AI models.
- Supports rare variants: Facilitates the interpretation of rare variants and sparse data inference within a unified architecture.
- Cross-species translation: Enhances research applicability through effective cross-species translation, addressing needs in clinical genomics, drug discovery, and toxicology.
Applications and Industries
- Mechanism-based drug repurposing platforms (pharmaceutical, biotech)
- Rare disease diagnostic support (clinical genomics, health systems)
- Model organism selection platforms (preclinical contract research organizations)
Contact:
To learn more about this technology, email partnerships@ornl.gov or call 865-574-1051.
