A network inference method for large-scale unsupervised identification of novel drug-drug interactions

Abstract

Characterizing interactions between drugs is important to avoid potentially harmful combinations, to reduce off-target effects of treatments and to fight antibiotic resistant pathogens, among others. Here we present a network inference algorithm to predict uncharacterized drug-drug interactions. Our algorithm takes, as its only input, sets of previously reported interactions, and does not require any pharmacological or biochemical information about the drugs, their targets or their mechanisms of action. Because the models we use are abstract, our approach can deal with adverse interactions, synergistic/antagonistic/suppressing interactions, or any other type of drug interaction. We show that our method is able to accurately predict interactions, both in exhaustive pairwise interaction data between small sets of drugs, and in large-scale databases. We also demonstrate that our algorithm can be used efficiently to discover interactions of new drugs as part of the drug discovery process.

Figure 1. Stochastic block models for the prediction of unknown drug interactions.

(A) Consider a hypothetical situation in which all of the interactions between drugs are known with the exception of the interaction between and , which is, in reality, antagonistic. There are many partitions of the drugs into groups. The partition in (B) has high explanatory power (low value of in Eqs. (5) and (6)), since most drug interactions between a pair of groups are of the same type. Therefore, the predictions of this partition have a large contribution to the estimation of the probability of the unknown interaction. Conversely, the partition depicted in (C) has little explanatory power (high value of ) and has a small contribution to the estimation of the probability of the unknown interaction.

Figure 2. Performance of drug interaction inference methods on exhaustive pair interaction data.

We test the algorithms against results of two experiments in which all pairwise interactions between a small set of drugs were tested: (A, C and E; interactions are synergistic, additive or antagonistic) and (B, D and F; interactions are synergistic, additive, antagonistic or suppressing). We simulate situations in which only a fraction of all interactions are observed, and then try to predict the unobserved interactions (repeated random sub-sampling validation). In each case, we measure the fraction of predictions that are exactly correct (A and B), as well as the fraction of predictions that deviate from the experimental observation by at most one level (C and D). For example, miss-predicting a synergistic interaction as additive is considered correct by the classification metric, but miss-predicting a synergistic interactions as antagonistic or suppressing (or vice versa), or an additive one as suppressing (or vice versa) is considered incorrect. Error bars indicate the standard error of the mean and are usually smaller than the symbols. (E and F) Relative improvement of the stochastic block model predictions over the neighbor-based predictions. If is the frequency of correct classification, we define the relative improvement as , where SBM and B stand for stochastic block model and baseline, respectively, and X stands for any other approach (neighbor-based or Prism-based).

Figure 3. Performance of drug interaction inference methods on an evolving database of major adverse drug interactions.

Left: Drugs.com database; right: DrugBank dataset. (A–B) Area under the receiver operating characteristic (AUROC) curve. For novel interactions the AUROC gives the probability that an interaction randomly chosen from those that were added to the first snapshot has a higher score than one randomly chosen from the set of interactions that were never added to the network. Similarly, for spurious interactions the AUROC gives the probability that an interaction randomly chosen from those that were removed from the first snapshot has a lower score than one randomly chosen from the set of interactions that were not removed from the network. (C–F) Sensitivity-specificity curves for novel (C–D) and spurious interactions (E–F). Sensitivity is defined as the ratio of true positives to all real positives (true positives plus false negatives). Specificity is defined as the ratio of true negatives to all real negatives (true negatives plus false positives).

Figure 4. Inference of drug interactions as part of the process of drug discovery and development.

For each of the two drugs ((A) acetophenazine and (B) cinacalcet) we simulate an iterative process in which a plausible interaction is suggested by the stochastic block model inference approach, the interaction is tested, and information is added to the network of known drug-drug interactions. The graphs display the number of true interactions discovered as a function of the number of experiments carried out. Green dots represent true interactions, whereas red dots represent drugs that were suggested as interaction candidates but turned out not to interact with the target drug. For acetophenazine, the 16 iterations we carry out are enough to discover 11 of the 15 interactions that are reported in DrugBank. For cinacalcet, we are able to uncover 8 of the 12 reported interactions. The gray region indicates the feasible region of discovery. Its upper bound corresponds to discovering all interactions without ever testing a drug that does not interact with the target drug; the lower bound corresponds to randomly exploring all possible interactions. In the lower bound, it takes around 100 experiments to uncover each interaction.