Fig 1.
Overview of the proposed approach.
(1) A PPI network is constructed starting from a list of disease proteins (DPs); then a list of target candidates (TPs) for drug synergy is obtained based on topological network properties; (2) A score function, called Topolgical Score of Drug Synergy (TSDS), assigns a score to each combination of TPs allowing the selection of significant multi-target combinations; (3) TP combinations are further augmented through the application of a data fusion approach. Here, the integration of several data sources [26] allows to obtain a list of known and predicted drug-target interactions; (4) The biological pathways related to disease progression are extracted; the pathways are represented with Boolean Networks (BNs); (5) BNs are simulated taking into account drug activities to understand biological pathways alterations through different pharmaceutical interventions. Finally, in vitro studies to validate the ability of the method to propose potential therapies can be carried on taking into account the results obtained from the previous phases.
Table 1.
Collection of data sources used for matrix tri-factorization, their size and number of edges.
Fig 2.
Predicting the drugs effect on biological pathways.
(1) Boolean modeling of KEGG pathways; (2) Modeling the disease nodes and the pharmacological actions; (3) Monte Carlo simulations of the drug combination actions; (4) Ranking of the drug efficacy and the disease proteins.
Table 2.
Conversion table of KEGG associations into Boolean rules.
Fig 3.
In the network the 43 DP seed nodes are highlighted in red while the 33 TP nodes are depicted by blue diamonds. The node size depends on the Bridging Centrality values as shown in the graph below the figure.
Fig 4.
Network constraints to select TP nodes.
In Fig 4(a) hubs are highlighted in pink. Note that these nodes are discarded as potential TPs. In Fig 4(b) orange nodes correspond to the bridging nodes, while in Fig 4(c) druggable nodes are depicted in dark green. The node size is proportional to its degree (i.e. number of neighbors).
Table 3.
List of network Target Proteins TP.
The column Freq. reports the protein frequency in the significant triplets.
Table 4.
Known and predicted drugs associated with significant TP nodes.
Table 5.
For each pathway, the number of nodes and edges of the related BN as well as the number of DPs and drug targets (for each of the drugs considered) present in the pathway networks are listed. In the table No TPs means that no drug targets were found in the pathway.
Fig 5.
Boolean Network of the Jak-STAT signalling pathway.
Fig 6.
An example of Odefy outputs obtained by simulating Imatinib administration in Jak-STAT signaling pathway.
Fig 7.
(a) PathEFFMC index for each simulated treatment in every pathway; (b) DugEFF(D) and the related noDrugEFF(D) for each drug combination; (c) EFFECT index for each simulated drug administration
Fig 8.
Evaluation of cell viability performed by treating MCF7 and MDA-MB-231 cell lines with different doses of Imatinib.
MCF7 cell line is taken as control, while MDA-MB-231 is used as representative of TNBC.
Fig 9.
Disease genes evaluations by treating disease and control cell lines with Imatinib.
Fig 10.
Evaluation of proliferation rate of TNBC cells (MDA-MB-231) and luminal-like breast cancer cells (MCF7).