Network-based prediction of drug combinations with quantum annealing

TL;DR

This paper presents a network-m medicine–informed quantum annealing framework to predict drug combinations by encoding Complementary Exposure as a QUBO and optimizing within a protein-protein interactome. By defining disease proximity metrics and a problem Hamiltonian with tunable parameters, the authors calibrate hyperparameters against a benchmark of validated combinations and use Simulated Quantum Annealing to sample low-energy candidate therapies. The approach yields disease-specific AP performance and demonstrates that low-energy states harbor validated combinations while also producing novel, plausible predictions when applied to larger drug sets. While not substituting experimental validation, the method provides a principled, scalable way to prioritize combinatorial therapies for experimental testing and future refinement.

Abstract

The systematic discovery of effective drug combinations is a challenging problem in modern pharmacology, driven by the combinatorial growth of potential pairings and dosage configurations. Network medicine, modeling diseases and drugs as interconnected modules of the human protein-protein interactome, has emerged as a new paradigm for understanding disease mechanisms and drug action. In this work, we propose a quantum annealing-based algorithm for identifying effective drug combinations. Underlying our approach is the biologically motivated principle of `Complementary Exposure', which posits that therapeutic drug combinations target distinct yet complementary regions of a disease module. We translate this into a quadratic unconstrained binary optimisation problem. We test our method for Diabetes Mellitus, Rheumatoid Arthritis, Asthma, and Brain Neoplasms, relying on experimentally validated drug combinations for these diseases. Our simulated quantum annealing experiments reveal that low-energy configurations align with biologically plausible combinations, demonstrating the algorithm's ability to generate novel predictions for drug combinations.

Figures (5)

• Figure 1: Schematic of the 'Complementary Exposure' principle. Two drugs, A (yellow) and B (blue), represented by their protein targets, exhibit shared nodes with the disease module (red) in the human protein-protein interactome graph. Both drug node sets target distinct, non-overlapping regions of the disease module (red). This leads to effective perturbation of the disease-associated network neighborhood as quantified by the network proximity measures $s_{AB}\geq0$ and $z_A,z_B<0$.
• Figure 2: Average Precision (AP) Landscape. Each subplot displays the AP values over a grid of hyperparameters $(\gamma, \beta)$ for the different diseases: (a) Diabetes Mellitus, (b) Rheumatoid Arthritis, (c) Asthma, and (d) Brain Neoplasms. The red dot in each plot indicates the optimal hyperparameter pair $(\gamma^*, \beta^*)$ that maximizes the AP metric, reflecting the best prioritization of validated drug combinations among low-energy configurations.
• Figure 3: Energy spectrum at the optimal hyperparameters Each subplot displays the top 15 lowest-energy configurations of the QUBO constructed from the initial candidate drug set at the disease-specific optimal hyperparameters $(\gamma^*, \beta^*)$. Solutions are ordered by increasing QUBO energy (lowest at the bottom) and are colour-coded to indicate whether the configuration matches a validated combination in the reference set $\mathcal{C}$ or not. The diseases represented are: (a) Diabetes Mellitus, (b) Rheumatoid Arthritis, (c) Asthma, and (d) Brain Neoplasms.
• Figure 4: Precision-Recall Curve at the optimal hyperparameters. Each subplot displays the Precision-Recall curve obtained by sweeping the ranked solutions of the QUBO constructed from the initial candidate drug set at the disease-specific optimal hyperparameters $(\gamma^*, \beta^*)$. The area under each curve corresponds to the Average Precision (AP) and quantifies the prioritization of validated drug combinations among the low-energy configurations. The diseases represented are: (a) Diabetes Mellitus, (b) Rheumatoid Arthritis, (c) Asthma, and (d) Brain Neoplasms.
• Figure 5: Sampled Simulated Quantum Annealing. Each subplot displays the empirical frequency distribution from $1024$ runs over the $10$ lowest QUBO energies obtained by running the SQA sampler on the enlarged candidate drug set $\mathcal{D}$ at the disease-specific optimal hyperparameters $(\gamma^*, \beta^*)$. Green markers indicate combinations found in $\mathcal{C}$ while the remaining configurations are taken as predicted candidate synergistic combinations. The diseases represented are: (a) Diabetes Mellitus, (b) Rheumatoid Arthritis, (c) Asthma, and (d) Brain Neoplasms.