Drug Research Meets Network Science: Where Are We?

Abstract

Network theory provides one of the most potent analysis tools for the study of complex systems. In this paper, we illustrate the network-based perspective in drug research and how it is coherent with the new paradigm of drug discovery. We first present data sources from which networks are built, then show some examples of how the networks can be used to investigate drug-related systems. A section is devoted to network-based inference applications, i.e., prediction methods based on interactomes, that can be used to identify putative drug-target interactions without resorting to 3D modeling. Finally, we present some aspects of Boolean networks dynamics, anticipating that it might become a very potent modeling framework to develop in silico screening protocols able to simulate phenotypic screening experiments. We conclude that network applications integrated with machine learning and 3D modeling methods will become an indispensable tool for computational drug discovery in the next years.

Figure 1

Exemplary CSN of PARP inhibitors. PARPs 1, 2, and 3 family inhibitors with a measured EC₅₀ were retrieved from CHEMBL. The pairwise chemical similarities between compounds were assessed by means of Tanimoto coefficient (Tc) values calculated for the ECFP4 fingerprints of the molecules generated by Canvas (Schrödinger, LLC, New York, NY, 2019). Pairs of inhibitors were connected only if their calculated Tc value exceeded the threshold value of 0.55. The chemical structures of the inhibitors are shown inside the nodes that are colored according to pEC₅₀ values ranging from red (lowest potency) to green (highest potency) and sized based on node degree from small (low degree) to large (high degree). Edges are weighted by Tc values from thin (Tc = 0.55) to thick (Tc = 1) width. The network was generated by means of Cytoscape version 3.7.2.

Figure 2

Drug–target network. The DTN was built from DrugBank version 5.1.5 retrieving the drug–target interactions between approved small molecule drugs and human protein targets. Drugs are represented as circle-shaped nodes, and protein targets are represented as diamond-shaped nodes. As shown in the inset, drugs are color-coded according to the first level anatomical therapeutic chemical (ATC) codes as reported in DrugBank. The nodes size accounts for the node degree from small (low degree) to large (high degree). Edges connect only drugs and targets nodes. The network was generated by means of Cytoscape version 3.7.2.

Figure 3

Hetionet version 1.0. (a) Metagraph showing the types of nodes used to build the network and the types of links defined to connect the nodes. A detailed description of the meaning of each link type as well as the sources of information used to collect the nodes and to draw the edges is reported in ref (85). (b) Visualization of the whole heterogeneous network. Nodes of the same type are grouped within circles, and links are colored by type. This Figure 3 is reproduced from Figure 1 of Himmelstein D. S.; Lizee A.; Hessler C.; Brueggeman L.; Chen S. L.; Hadley D.; Green A.; Khankhanian P.; Baranzini S. E.; 2017; eLife (ref (85)), published under the Creative Commons Attribution 4.0 International Public License (CC BY 4.0; https://creativecommons.org/licenses/by/4.0/).

Figure 4

Boolean network dynamics. (a) Nodes are colored red or gray based on their “on” or “off” state, respectively. The stepwise evolution of the interactions between nodes determines some sequential steps that are calculated based on a set of rules. The stable state of the network at time step t = S represents an attractor. (b) Attractor landscape. Gray circles represent network states, colored circles represent attractor states. The landscape contains all the possible network states. The sets of states that converge toward an attractor form the basin of that attractor (colored areas). (c) The basins of attractors can be associated with cell phenotypes, and the gene states of the attractors determine the nature of the phenotype.