Network-based prediction of drug combinations.
Cheng F, Kovács IA, Barabási AL. · 2019
Read Network-based prediction of drug combinations. by Cheng F, Kovács IA, Barabási AL. (2019) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6416394.txt.
Introduction
Cheng and co-authors present a network method in Nature Communications for finding drug pairs. Pairs of drugs can work better and harm less than one drug alone. But the number of possible pairs is far too large to test one by one.
Their method maps drug targets and disease proteins on the human protein network. Six classes of drug-drug-disease patterns emerged from that map. Only one class matched pairs known to work in patients. In that class, both drugs hit the disease area but in separate parts of it. The rule may help screen pairs before costly trials.
Catalog record
| CREATOR | Cheng F, Kovács IA, Barabási AL. |
|---|---|
| YEAR | 2019 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Network-based prediction of drug combinations By Feixiong Cheng, István A Kovács, Albert-László Barabási Published in: Nature Communications DOI: 10.1038/s41467-019-09186-x ABSTRACT Abstract Drug combinations, offering increased therapeutic efficacy and reduced toxicity, play an important role in treating multiple complex diseases. Yet, our ability to identify and validate effective combinations is limited by a combinatorial explosion, driven by both the large number of drug pairs as well as dosage combinations. Here we propose a network-based methodology to identify clinically efficacious drug combinations for specific diseases. By quantifying the network-based relationship between drug targets and disease proteins in the human protein–protein interactome, we show the existence of six distinct classes of drug–drug–disease combinations. Relying on approved drug combinations for hypertension and cancer, we find that only one of the six classes correlates with therapeutic effects: if the targets of the drugs both hit disease module, but target separate neighborhoods. This finding allows us to identify and validate antihypertensive combinations, offering a generic, powerful network methodology to identify efficacious combination therapies in drug development. Subject terms: Networks and systems biology, Network topology, Target identification, Biochemical networks INTRODUCTION Combination therapy, the use of multiple drugs to improve clinical outcomes, has multiple advantages compared to monotherapy1,2: it offers higher efficacies or, through lower individual dosage, it can reduce the risk of adverse effects3. Consequently, combination therapies are widely used in the treatment of multiple complex diseases, from hypertension 4 to cancer5 and infectious diseases6,7. However, the systematic identification of drug combinations that simultaneously offer high clinical efficacy and low toxicity is often driven by intuition and experience rather than established principles. There is a pressing need, therefore, for novel methodologies to facilitate the discovery of multicomponent therapy. One approach is the systematic high-throughput testing of pairwise drug combinations, which, however, faces a combinatorial challenge: for 1000 U.S. Food and Drug Administration (FDA)-approved drugs, there are 499,500 possible pairwise combinations that should be tested over approximately 3000 human diseases and multiple dosage combinations5,6,8,9. We are, therefore, far from even a cursory exploration of the vast number of possible combinations with potentially positive clinical outcomes. To be sure, several machine learning-based “black-box” models have been developed to predict drug combinations10–12, offering a modest increase in accuracy10 over random guesses11. We lack, however, predictive, mechanism-driven, network-medicine-based approaches to predict efficacious drug combinations. Network-based approaches have already offered a promising framework to identify novel insights to accelerate drug discovery13, helping us quantify both disease–disease14 and drug–disease15,16 relationships. These methodological advances have raised the possibility of moving beyond the “one-drug, one-target” paradigm and exploring the “multiple-drugs, multiple-targets” possibilities offered by aiming at simultaneously modulating multiple disease proteins within the same disease module, while minimizing toxicity profiles17–21. In this study, we quantify the relationship between drug targets and disease proteins in the human protein–protein interactome, leading to a rational, network-based drug combination design strategy. RESULTS Network-based proximity measure of drug–drug relationships Disease proteins are not scattered randomly in the interactome, but tend to form localized neighborhoods, known as disease modules14. An efficient way to capture network proximity between a drug (X) and a disease (Y) is by the z-score (z=d-μσ), which relies on the shortest path lengths d(x, y) between drug targets (x) and disease proteins (y). [formula: 1dX,Y=1Y∑y∈Yminx∈Xd(x,y)] The z-score is obtained by comparing the observed distance to a reference distance distribution between a randomly selected group of proteins of matching size and degree distribution as the disease proteins and drug targets in the human interactome. The z-score is applicable when the reference distribution is well described by a Gaussian (Supplementary Note 1) and was successfully used to identify well-known and new drug–disease relationships in the human interactome for monotherapy15. Here, we hypothesize that exploring the network-based relationship between two drugs and their targets, and the disease proteins in the disease module would help clarify the mechanism-of-action of effective drug combinations while minimizing adverse effects (Supplementary Fig. 1). To test our hypothesis, we assembled 243,603 experimentally confirmed protein–protein interactions (PPIs) connecting 16,677 unique proteins from five data sources (Supplementary Note 2 and Supplementary Data 1). We also compiled 1978 FDA-approved or clinically investigational drugs that have at least two experimentally reported targets by pooling the high-quality drug–target binding affinity profiles from six data sources (Supplementary Data 2). In order to fully characterize the mutual relationship of two drugs and a disease module, we need a network-based proximity measure between the two drugs’ targets, as well. Although it has not been tested for this purpose, in principle, we could rely on the z-score for this task. However, in contrast to the relatively large disease modules, each drug has only a small number of experimentally reported targets (on average 3, Supplementary Fig. 2). Therefore, the randomization procedure is not producing a Gaussian distribution as described in our previous study15, limiting the applicability of the z-score. Indeed, we find that the z-score cannot discriminate FDA-approved pairwise combinations or clinically reported adverse drug interactions from random drug pairs (Supplementary Fig. 3). Instead of relying on randomization, therefore, we measure the network proximity of drug–target modules A, B as reflected in their target localizations using the recently introduced separation measure14: [formula: 2sAB≡dAB-dAA+dBB2] which compares the mean shortest distance within the interactome between the targets of each drug, 〈dAA〉 and 〈dBB〉, to the mean shortest distance 〈dAB〉 between A–B target pairs (Fig. 1a). In 〈dAB〉, targets associated with both drugs A and B have a zero distance by definition. For sAB < 0, the targets of the two drugs are located in the same network neighborhood (Fig. 1b), while for sAB ≥ 0, the two drug targets are topologically separated (Fig. 1c). For example, imatinib (I) is an FDA-approved agent for the treatment of chronic myeloid leukemia22. Figure 1a, b shows that imatinib’s targets are in the same network neighborhood as the targets of tandutinib (T), a FMS-like tyrosine kinase 3 (FLT3) inhibitor under phase III trial for treating acute myeloid leukemia23; consequently the separation score between their targets is negative, sIT = −0.35. However, the targets of natalizumab (N), an FDA-approved monoclonal antibody for treating multiple sclerosis24, are in a topologically distinct neighborhood from the targets of imatinib and tandutinib, having a positive sIN = 0.59 and sTN = 0.49, respectively (Fig. 1c). [Figure: Network-based model of drug–drug relationship. a A subnetwork of the human interactome illustrating the network-based relationship between drug targets associated with three drugs (imatinib [I], tandutinib [T], and natalizumab [N]). b, c The definition of drug pairs that are topologically overlapping (sAB < 0, b) or topologically separated (sAB ≥ 0, c). d–j The interplay between topological separation of drug pairs and five types of drug profiles: drug–drug chemical similarity (d); drug target-encoding gene co-expression pattern across human tissues (e); drug target protein sequence similarity (f). Using the Gene Ontology (GO) annotations, we determine for each drug how similar its associated target-encoding genes are in terms of their biological processes (g), cellular component (h), and molecular function (i); and clinical similarity (j) of drug pairs derived from Anatomical Therapeutic Chemical Classification Systems codes (see Methods). Overlapping drug pairs are highlighted in orange (sAB < 0); topologically separated drug pairs are highlighted in blue (sAB ≥ 0)] For a network-based approach to drug