Chapter7 : Network-Driven Drug Discovery

reading notes of《Artificial Intelligence in Drug Design》

---

1.Motivations

• A major problem, arguably the major problem, facing drug discovery is the complexity of human biology.
• The complexity of human biology hampers drug discovery at multiple levels:
• the identification of biological processes involved in disease
• understanding the molecular basis of those processes
• identification of molecular intervention points
• the design and development of agents that can specifically trigger those interventions
• The need for novel, systems-based approaches within early stage drug discovery is highlighted by a number of observations:
  • a large number of late stage clinical failure are due to targeting of incorrect or ineffective
  • the search for treatments for complex diseases has had limited success
  • Post-genomic molecular profiling technologies have enabled the state of a cell to be measured system-wide across multiple modalities. However, the ability to leverage these data to elucidate the molecular basis of disease, and produce actionable insights for drug discovery, has proved more elusive
  • The complexity of human biology is reflected in the failure of pre-clinical assays to translate from model systems to humans. The desire to improve translatability has led to the development and increased use of more complex assays, very often using human tissue, that attempt to reflect better the critical elements of the disease
  • Recent advances in personalized medicine have illustrated that patients’ disease may segment at a process level despite sharing a superficially common phenotype. Specific subgroups of a larger patient population can be defined on the basis of similar molecular underpinnings, and those molecular underpinnings are distinct across the subgroups. The existence of multiple, distinct mechanistic bases for the same disease is rarely considered during the early stages of drug discovery and is another prominent factor in the lack of translatability of experimental results from model systems
• The approach combines network-biology-based computational approaches with experimental cell-based complex phenotypic assays. This combined computational and experimental approach aims to incorporate the advantages of each to build a “best of both worlds” process.

2.Network Biology and Pharmacology

• The networks of intra- and intercellular interactions, and the dynamic processes operating on them, can be viewed as the molecular substrate for biological function in cells, tissues and organs; networks act as a mechanistic bridge between the molecular level and the phenotypic
• A cellular process that is resilient to perturbation and robust to random failure of its individual molecular components. Such processes can in turn interact with other processes at multiple levels to produce more complex phenotypes.
• All such networks contain sets of key proteins that can trigger state changes which alter their phenotypic behavior in response to relevant external signals. For any external intervention to have a significant effect on a functional subnetwork the multiple downstream modulations arising from target engagement must ultimately influence the key proteins in this subnetwork if the aim is to modulate the biological function of interest and thus alter phenotypic behavior.
• Generally, it is the downstream consequences of the manipulation, “the effector signature” that can be said to be the true pharmacological intervention which modulates the operation of the network. To understand drug effects, we need to consider the impact of such signatures on the networks in which they act. This “action at a distance” concept is illustrated in Fig.1.

3.Implications for Drug Discovery

• This conceptual shift from searching for agents that can affect phenotype to searching for agents that can perturb the underlying, network-based mechanisms has a number of implications that must be considered in order to develop such a systems-based approach to drug discovery successfully.

3.1.Network Identification

• The first stage in a network-driven approach to drug discovery is to formulate intervention strategy hypotheses.
• In such a disease context, given a suitable collection of molecular profiling measurements, the network modules associated with specific mechanistic subgroups can be identified and used to identify subgroup-specific biological processes that can subsequently drive discovery.

3.2.Functional Redundancy

• A key property of complex biological systems is the seemingly paradoxical coexistence of phenotypic robustness and phenotypic adaptability.
• The mechanistic basis of this coexistence of robustness and adaptability has been proposed to be degeneracy, or functional redundancy, where the same biological function can be carried out by structurally different mechanisms.
• Mechanisms underlying biological robustness can also confer stability to a diseased system resulting in therapeutic interventions with lower than expected efficacy or adaptation in the face of drug application.
• Rather than aiming for focused, accurate representations of the canonical pathways and networks, the construction processes should aim additionally to capture degenerate structures reflected in non-canonical pathways, pathway–pathway interactions, and the like.

3.3.Network Perturbation

• This being the case, a key consideration must be the identification of network nodes (e.g., proteins) that, when perturbed as a unified set, will result in significant network modulation.
• Perturbation of networks can be assessed using many different measures but usually involves quantification of changes in some structural statistic of the network following an intervention (e.g., deletion of a node).
• Focusing on the structural aspects of networks, while ignoring dynamic processes that operate on those networks, is a simplification but is nevertheless a useful and a pragmatic one, especially since network dynamics are significantly constrained by network structure.

3.4.Action at a Distance

• Drugs can have an “action at a distance” via signaling or metabolic pathways that convert drug binding events into downstream functional effects; and this ‘action at a distance’ must be considered when identifying possible targets and/or therapeutic agents.

4.Network-Driven Drug Discovery

• The network-driven discovery process was deliberately designed to be statistical in nature to address the well-known issues around noisiness and irreproducibility of biological data.
• The key outputs of the network-driven approach can be summarized as follows:
  • Mechanistic hypothesis generation
  • Active compound identification and target deconvolution
  • Target identification and validation
  • Human genetic evidence

5.Validation

• The network-driven approach to early stage discovery described in this chapter has been validated through its deployment in a number of real-world drug discovery programs across a range of complex biological and therapeutic settings.

5.1.COVID-19

• Building on previous anti-viral work we utilized our network biology platform to try to B that either singly or in combination could be repurposed rapidly to treat COVID-19.

6.Summary

• The chapter has outlined a conceptual framework in which network biology is the connecting bridge between individual molecular entities and emergent functional phenotype.
• We have also described the basic outline and validation of a network-biology-driven approach to drug discovery that, while disease agnostic, is particularly applicable in complex disease and which additionally permits the interpretation and deployment of complex and hitherto poorly leveraged genetic data such as that derived from population genetics.