Drug-disease networks and drug repurposing

TL;DR

This paper reframes drug repurposing as a link-prediction task on a large bipartite drug–disease network and demonstrates that topology-only methods can identify high-quality repurposing candidates. It constructs the network from public resources (2620 drugs, 1669 diseases, 8946 known indications) and evaluates a broad spectrum of link-prediction techniques, including graph embeddings, matrix factorization, and stochastic block models. The results show AUROC values exceeding 0.95 for strong methods, with SVD and NNMF delivering top precision (AUPR ≈ 20% and top-100 precision ≈ 70%), while the microcanonical SBM attains the best AUROC around 0.967 when extrapolated to the full network. The study provides strong evidence that network-based predictions can substantially reduce the search space for drug repurposing, offers bounds on missing edges and false positives, and supplies data/code for reproducibility and future improvements.

Abstract

Repurposing existing drugs to treat new diseases is a cost-effective alternative to de novo drug development, but there are millions of potential drug-disease combinations to be considered with only a small fraction being viable. In silico predictions of drug-disease associations can be invaluable for reducing the size of the search space. In this work we present a novel network of drugs and the diseases they treat, compiled using a combination of existing textual and machine-readable databases, natural-language processing tools, and hand curation, and analyze it using network-based link prediction methods to identify potential drug-disease combinations. We measure the efficacy of these methods using cross-validation tests and find that several methods, particularly those based on graph embedding and network model fitting, achieve impressive prediction performance, significantly better than previous approaches, with area under the ROC curve above 0.95 and average precision almost a thousand times better than chance.

Figures (4)

• Figure 1: A bipartite network of drugs and diseases of the type considered here. Nodes in the network are of two types---drugs and diseases---and edges connect only nodes of unlike types to indicate which drugs are indicated for treatment of which diseases. The network is assumed to be incomplete, so some edges that should be present are missing from the data (dashed lines). Our goal is to identify these.
• Figure 2: Visualization of the complete network of drugs, diseases, and their therapeutic interactions. Drug nodes are shown in blue and disease nodes in yellow.
• Figure 3: A visualization of the network of drugs and diseases. In this figure, we show (circles) a subset of the diseases in our network, those falling in the ten most common disease categories, as labeled, along with the shared drugs that connect them together (gray squares).
• Figure 4: (a) Receiver operating characteristic (ROC) curves for seven of the best performing algorithms. The dashed diagonal line represents the expected performance of a no-skill (random) classifier. Inset: Area under the curve (AUROC) for the microcanonical SBM for various fractions of edges removed. (b) Precision/recall curves for the same selection of algorithms. Colors are the same as in (a). Inset: Top-100 precision and area under the prediction/recall curve (AUPR) for each algorithm.