Brief comments on “Perturbation response scanning of drug-target networks: Drug repurposing for multiple sclerosis”

The recent Nobel prizes in Physics to Giorgio Parisi, Geoffrey Hinton, and John Hopfield, officially proclaimed a deep epistemological change: the unity of different sciences is no more considered to stem from the fact that ‘any entity is made by the same basic bricks’ but on the recognition than ‘any entity can be represented as a set of mutually interacting parts’. That is to say that any system [1] can be formalized as a ‘network of interactions among its elements’.

The cross-disciplinary portability of such relational models defines these approaches as ‘theory-free’, in the sense of independence from constitutive, field-specific, microscopic-level theories, as actually happens in thermodynamics [2].

The work by Lu et al. [3] is a perfect application of this new scientific approach to the goal of finding a drug candidate to repurpose toward the cure of multiple sclerosis (MS).

The multi-scale character of biological organization allows the authors to generate different formalization of the ‘wiring structure’ of the system at hand, i.e., to construct interaction graphs correspondent to different organization layers: similarity among diseases in terms of affected genes, similarities among drugs in term of hypothetical targets, protein-protein interaction graphs, and drug-target networks. This strategy leverages the consilience among these different views of MS fostering a rational strategy of drug repurposing by individuating the most promising therapeutic candidates among drugs already proven to be effective in related disease.

This consilience among different views was assessed by both static and dynamic approaches. The static approaches leveraged the topological features of the different networks in order to check for the ‘reachability’ of specific therapeutically relevant nodes as well as the identification of ‘signal-spreaders’ (nodes that have the ability, when perturbed, to pass the signal across the network) or the nodes acting as module connectors (nodes that are crucial to keep together different parts of the network).

The dynamic approach consisted in the perturbation response analysis (perturbation response scanning (PRS)) method: this is a strategy developed in the frame of molecular dynamics that, thanks to the universality of topological laws can be applied to any network, irrespective of the nature of the nodes. The idea at the basis of PRS is both simple and general. The action of a drug can be equated to a force (perturbation) applied to a node: the effect of the drug on the system corresponds to the spreading of the signal across the network that in turn equated the way in which the network system dissipates the excess of energy due to the perturbation [4].

The analyzed networks described two complementary spaces and were both analyzed in terms of PRS and nodes topological role.

• i)Phenotypic space: a comorbidity space of 845 MS-related disease genes was the basis for the selection of 100 shared genes, which in turn, through mapping onto the human protein-protein interaction network (PPIN), generate the seed PPIN made of 80 interconnected nodes.
• ii)Drug-target space: Following the same strategy described for phenotypic space, the superposition among the targets of effective drugs for the diseases, having the maximal consilience with MS in phenotypic space, was exploited in order to generate a priority score for wet-lab testing.

This is, in my opinion, the most striking methodological novelty of the work [3]: the application of a physically motivated method to the study networks not having a direct physical counterpart (the network comes from experimental results coming from literature) implies a sort of ‘statistical mechanics of data’ [5] that, at least in my opinion (even thanks to the rapid evolution of physics-inspired machine learning approaches [2]), have the potential to become the next ‘big revolution in science’.

But the methodological novelty does not stand alone: the authors picked up the most promising drug candidate suggested by their computational approach and performed wet-lab in vivo tests giving a robust (albeit preliminary) proof of the proposed strategy.

All in all, I consider this paper as a seminal work for the development of a new strategy in pharmacological research that promises to foster a new era in pharmacological research.