¹Institut für Infektionsmedizin und Krankenhaushygiene, Universitätsklinikum Jena, Jena, Germany
²Institut Pasteur, Lyon, France, Lyon, France
³Mattek Europe, Bratislava, Slovakia, Bratislava, Slovakia
⁴Ruprecht Karls University, Heidelberg, Germany
⁵Technical University of Munich, Munich, Germany
⁶Institut de Recherche en Infectiologie de Montpellier (IRIM), CNRS, Montpellier, France

Text

Introduction: The emergence and re-emergence of viral pathogens have the potential to cause life-threatening epidemics, particularly in our increasingly globalized and ecologically challenging world. A critical aspect of pandemic preparedness is the development of effective antiviral therapies, particularly broad-spectrum antivirals that target host dependency factors (HDFs), thereby reducing the risk of resistance development. The EU-funded project APPEAL (Antivirus Pandemic Preparedness EuropeAn pLatform) integrates computational and experimental approaches to identify broad-spectrum antiviral HDFs along with corresponding drugs. The aim of APPEAL is to implement a drug selection pipeline. This work is one part of the pipeline that helps selecting promising drug candidates for our experimental collaboration partners.

Methods: Given a list of HDFs, the PPI-network extracted from BioGRID (https://thebiogrid.org/) and STRING (https://string-db.org/), and different drug databases (Drugbank (https://go.drugbank.com/), ChEMBL (https://www.ebi.ac.uk/chembl/), TTD (https://idrblab.net/ttd/), PharmGKB (https://www.pharmgkb.org/), and BindingDB (https://www.bindingdb.org/)), we use a modified version of the method of Cheng et al. [1] to assemble a list of drugs that fit best to the given HDFs as targets. In this work the exact shortest path lengths between drugs and targets of interest are used and compared to random networks. We instead are not interested in longer distances and therefore introduce a distance-metric within the PPI-network distinguishing only three cases:

• proteins that are an HDF have distance 0,
• proteins that are not an HDF but directly connected to one via an edge in the PPI-network have distance 1,
• all other proteins have distance 2 (even if the minimum shortest path to an HDF is higher).

The method minimizes the distances of the targets of a drug to the HDFs, in comparison with distances calculated from randomly selected proteins with similar degree. With this method. we obtain drugs that are locally (regarding the network) close to our HDF-list. In addition, we obtain a rather different order of the drugs if we put their targets into the three distance groups and perform tests for enrichment of HDFs. This latter approach rewards drugs that are globally reflecting our HDF-list. The final ordering is then a combination of both the local and the global one.

In a second step we extend the method to find also best combinations of drugs. Since the total number of possible combinations explodes very quickly we use a greedy like method for this. We take the best 100 combinations of length k, combine them with all possible drugs to obtain combinations of length k+1, take again the best 100 and repeat with k+1.

Results/conclusion: The resulting lists of ordered drugs were compared with the literature evidencing several top ranking hits as inhibitors of known HDFs of viruses of our concern, but also so far not known HDFs. An example for a top hit is “tiludronic acid”, this drug is algorithmicly close to our HDFs. Of course the question of applicability still remains. Lists of drugs and drug-combinations are passed to our experimental collaboration partners within APPEAL to follow up with experimental validation.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.

References

[1] Cheng F, Kovács IA, Barabási AL. Network-based prediction of drug combinations. Nat Commun. 2019 Mar 13;10(1):1197. DOI: 10.1038/s41467-019-09186-x [Erratum in: Nat Commun. 2019 Apr 15;10(1):1806. DOI: 10.1038/s41467-019-09692-y]