A Novel Framework for Comparing Combination Therapy Outcomes Using Mechanistic Graph Models

Background: Predicting the efficacy of combination therapies is a critical challenge in clinical decision-making, particularly for diseases requiring multi-drug regimens. Traditional evidence synthesis methods, such as component network meta-analysis (cNMA), often face parameter explosion and limited interpretability, especially when modeling interaction effects between components. Objective: This article introduces a general Efficacy Comparison Framework (ECF), a mechanistically grounded system for predicting combination therapy outcomes. ECF integrates biological pathway-based abstractions with expert knowledge, optimized with quasi-rules derived from clinical trial data to overcome the limitations of traditional methods. Methods: ECF employs a disease pathogenesis graph to encode domain knowledge, reducing the parameter space through mechanistic functions and sparse network structures. Optimization may be performed using a loss function inspired by the Thurstone-Mosteller model, focusing on pairwise regimen comparisons. A pilot study was conducted for acne vulgaris to evaluate ECF's ability in both tested and untested comparisons. Results: In the acne vulgaris case study, the ECF-based model achieved 76% accuracy in predicting both tested and untested regimen outcomes, demonstrating statistically comparable performance across clinical trial data and expert dermatologist consensus (p = 0.977). The agreement between ECF and expert predictions was within the range of inter-expert agreement. Discussion: ECF aligns with recent advancements in network science and synergy prediction, leveraging principles of complementary targeting and biological plausibility. Its use of disease pathogenesis graphs offers a more interpretable and scalable alternative to existing models reliant on chemical similarity or protein-protein interaction (PPI) topology.