Causal knowledge graph analysis identifies adverse drug effects (2505.06949v1)

Abstract: Knowledge graphs and structural causal models have each proven valuable for organizing biomedical knowledge and estimating causal effects, but remain largely disconnected: knowledge graphs encode qualitative relationships focusing on facts and deductive reasoning without formal probabilistic semantics, while causal models lack integration with background knowledge in knowledge graphs and have no access to the deductive reasoning capabilities that knowledge graphs provide. To bridge this gap, we introduce a novel formulation of Causal Knowledge Graphs (CKGs) which extend knowledge graphs with formal causal semantics, preserving their deductive capabilities while enabling principled causal inference. CKGs support deconfounding via explicitly marked causal edges and facilitate hypothesis formulation aligned with both encoded and entailed background knowledge. We constructed a Drug-Disease CKG (DD-CKG) integrating disease progression pathways, drug indications, side-effects, and hierarchical disease classification to enable automated large-scale mediation analysis. Applied to UK Biobank and MIMIC-IV cohorts, we tested whether drugs mediate effects between indications and downstream disease progression, adjusting for confounders inferred from the DD-CKG. Our approach successfully reproduced known adverse drug reactions with high precision while identifying previously undocumented significant candidate adverse effects. Further validation through side effect similarity analysis demonstrated that combining our predicted drug effects with established databases significantly improves the prediction of shared drug indications, supporting the clinical relevance of our novel findings. These results demonstrate that our methodology provides a generalizable, knowledge-driven framework for scalable causal inference.

Analysis of Adverse Drug Effects Using Causal Knowledge Graphs

The paper introduces a novel methodology for identifying adverse drug effects through the integration of knowledge graphs with structural causal models, resulting in the formation of Causal Knowledge Graphs (CKGs). This integration addresses key limitations in existing computational models used in biomedical domains, providing a framework for scalable causal inference that can enhance drug safety monitoring.

Novel Framework: Causal Knowledge Graphs

CKGs utilize the structural nature of knowledge graphs while embedding formal causal semantics that facilitate both deductive and causal reasoning. By marking causal edges explicitly in the graph, CKGs can support deconfounding and hypothesis generation aligned with both encoded and entailed knowledge. This is particularly crucial for biomedical applications where complex interactions among various factors need to be accurately modeled.

The researchers constructed a Drug-Disease Causal Knowledge Graph (DD-CKG), encompassing disease progression pathways, drug indications, adverse effects, and hierarchical disease classifications. Utilization of cohorts from UK Biobank and MIMIC-IV allowed automated large-scale mediation analysis to determine whether drugs mediate effects between their indications and disease progression pathways. The DD-CKG effectively reproduced known adverse drug reactions and proposed novel candidate effects not previously documented.

Key Findings

1. Replication of Known ADRs: The approach demonstrated high precision in replicating established adverse drug reactions, supporting the validity of the causal inference mechanism integrated within the DD-CKG.
2. Identification of Candidate ADRs: The methodology identified significant, previously undocumented candidate adverse effects, adding value to the drug safety surveillance paradigms.
3. Improvement in Predicting Drug Indications: By combining predicted drug effects with existing databases, the methodology improved prediction of shared drug indications, highlighting the clinical relevance of its novel findings.
4. Confounding Control: The integration of CKGs enabled effective control for confounders, which are often challenging to manage in traditional data-driven models without explicit knowledge representation.

Implications and Future Directions

The implications of this research are significant for post-marketing surveillance of drugs. By providing a framework that can scale and adapt to the integration of extensive observational data sources with structured biomedical knowledge, CKGs propose a pathway towards more holistic and accurate pharmacovigilance systems. The strong numerical results demonstrated in the paper reinforce the potential for future developments in AI that may leverage the structure and causality of graphs for improved inference accuracy.

Theoretical implications suggest broader applications beyond pharmacovigilance into any domain where causal inference intersects with structured domain knowledge, such as genomics, systems biology, and personalized medicine. As the field evolves, integrating causal graph semantics with large-scale machine learning models could unlock new possibilities for predicting complex biological interactions and disease mechanisms.

This paper highlights the importance of continued exploration into methodologies that harness the power of structured knowledge combined with causality—in paving the path for more sophisticated biomedical informatics tools capable of addressing challenges in drug safety and therapeutic efficacy.