Structure and dynamics of molecular networks: A novel paradigm of drug discovery. A comprehensive review

Peter Csermely; Tamas Korcsmaros; Huba J. M. Kiss; Gabor London; Ruth Nussinov

Source-linked AI summary

Structure and dynamics of molecular networks: A novel paradigm of drug discovery. A comprehensive review

Peter Csermely, Tamas Korcsmaros, Huba J. M. Kiss, Gabor London, Ruth Nussinov

arXiv:1210.0330v3Oct 1, 2012q-bio.MNcond-mat.dis-nncs.SInlin.AOphysics.bio-ph

TL;DR

Drug development is costly and inefficient despite extensive molecular screening, while disease mechanisms and translational evidence remain complex. This review synthesizes network topology and dynamics across molecular and disease data to guide target discovery, hit identification, lead optimization, and safety assessment. It proposes central-hit and network-influence strategies, while emphasizing data-quality limits and the need for further comparative studies.

Problem
Drug development remains costly and inefficient, and network methods depend on incomplete or unreliable disease and system-dynamics data.
Method
The review comprehensively assesses network topology and dynamics and synthesizes applications across disease, molecular, target, biomarker, safety, and resistance networks.
Results
The review proposes central-hit targeting for flexible infectious-agent or cancer networks and network-influence targeting for rigid complex-disease networks.
Takeaways & Limitations
Network analysis may support disease-specific target identification, molecule selection, efficacy optimization, and reduction of side-effects, toxicity, and resistance.
Takeaways & Limitations
Network methods are constrained by input-data quality, limited high-quality system-dynamics datasets, and the need for comparative studies establishing where they help most efficiently.

Abstract

from arXiv · show

Despite considerable progress in genome- and proteome-based high-throughput screening methods and in rational drug design, the increase in approved drugs in the past decade did not match the increase of drug development costs. Network description and analysis not only give a systems-level understanding of drug action and disease complexity, but can also help to improve the efficiency of drug design. We give a comprehensive assessment of the analytical tools of network topology and dynamics. The state-of-the-art use of chemical similarity, protein structure, protein-protein interaction, signaling, genetic interaction and metabolic networks in the discovery of drug targets is summarized. We propose that network targeting follows two basic strategies. The central hit strategy selectively targets central nodes/edges of the flexible networks of infectious agents or cancer cells to kill them. The network influence strategy works against other diseases, where an efficient reconfiguration of rigid networks needs to be achieved by targeting the neighbors of central nodes or edges. It is shown how network techniques can help in the identification of single-target, edgetic, multi-target and allo-network drug target candidates. We review the recent boom in network methods helping hit identification, lead selection optimizing drug efficacy, as well as minimizing side-effects and drug toxicity. Successful network-based drug development strategies are shown through the examples of infections, cancer, metabolic diseases, neurodegenerative diseases and aging. Summarizing more than 1200 references we suggest an optimized protocol of network-aided drug development, and provide a list of systems-level hallmarks of drug quality. Finally, we highlight network-related drug development trends helping to achieve these hallmarks by a cohesive, global approach.

1. Introduction

Drug development faces rising costs, limited approvals, complex disease biology, and unreliable translational evidence. The review presents network description and analysis as a systems-level approach for organizing disease complexity and identifying drug-design opportunities.

Network analysis addresses biological complexity by integrating knowledge across molecular and disease relationships, supporting target, biomarker, hit, and lead-development tasks.
12 to 15 years and as much as 1 billion USD may be required to bring a single drug to market.
1.1. Drug design as an area requiring a complex approach: Despite higher R&D investment, FDA approvals remained at 20 to 30 new molecular entities annually, with only 6 to 17 potentially substantial advances.
1.1. Drug design as an area requiring a complex approach: Target selection is a major productivity bottleneck because chemically approachable targets can advance despite poor target quality, while later-stage leads may show toxicity or side-effects.
Preclinical evidence is often difficult to reproduce: Amgen reproduced 11% of 53 anticancer studies, while Bayer reproduced 25% of previously published studies.
1.3.2. The human disease network: Disease-related network patterns include tissue specificity, overlapping disease modules, and differing interactome positions for orphan, cancer, and common diseases.

2. An inventory of network analysis tools helping drug design

The review defines network construction carefully and surveys sampling, prediction, and reverse-engineering tools for extracting reliable molecular-network information relevant to drug design.

Network definition: Network analysis begins by defining nodes, edges, weights, directions, thresholds, and observation windows because these choices determine the represented interactions.Edges may encode physical or functional interactions, while weights represent interaction intensity, probability, or affinity.
Sampling and incompleteness: Incomplete biological data contain undetected interactions and false positives, making representative sampling and network-quality assessment essential.Networks are heterogeneous, so methods can test representativeness, extrapolate partial data, and filter while preserving topology and weight distributions.
Prediction: Missing-edge and missing-node prediction supports network reliability assessment, discovery of disease-protein interactions, and extension of drug-target networks.Prediction methods use node properties, neighborhood similarity, model-network comparisons, and iterative procedures, with performance varying across network regions.
Prediction: Drug-target edge prediction can discover new candidates and reposition existing drugs by combining expression, genotype, DNA–protein, and protein–protein interaction data.Combining data sources may improve precision, although directed, weighted, signed, or colored edges remain more difficult to predict.
Prediction: Network dynamics limit edge predictability because unexpected interactions may be mistaken for spurious edges and dense cores are more predictable than peripheries.Changing hub neighborhoods illustrate how network dynamics contribute to inherent unpredictability.
Reverse engineering: Reverse engineering has reconstructed drug-affected pathways and identified gene-regulatory and signaling networks from molecular measurements.The reviewed applications include transcriptome, phosphorome, and signaling-network data.

2.3. Key segments of network structure

The review distinguishes local, mesoscopic, and global network positions that can mediate information flow or perturbations, while balancing therapeutic influence against toxicity.

Local topology: Hubs have many more neighbors than average and can rapidly disrupt information transfer when attacked, but essential hubs may increase toxicity.Approved-drug targets tend to be more connected than peripheral nodes but less connected than hubs, while cancer proteins often have many interaction partners.
Local topology: Weighted and directed connectivity distinguishes hubs with equal versus uneven edge strengths and separates source-hubs from sink-hubs.These distinctions refine simple degree-based target selection.
Mesoscopic structure: Motifs, graphlets, modules, overlaps, and network skeletons provide structural representations linked to protein function, perturbation flow, and drug-target selection.Modules encode cellular functions; overlapping nodes transmit perturbations, while skeletons comprise interconnected high-centrality nodes.
Modules and bridges: Bridges connect neighboring modules and may offer lower-toxicity targets by disrupting information flow between functional network modules.Bridges usually have fewer neighbors than hubs and are independently regulated from nodes in the connected modules.
Modules and bottlenecks: Bottlenecks occupy unique inter-modular positions through which almost all network information passes, making them more effective but potentially more disruptive targets than bridges.Their use is therefore mainly restricted to anti-infectious and cancer-specific therapies.
Global topology and targeting: For infectious and cancer therapies, network integrity helps identify essential proteins whose disruption can destroy parasite or malignant-cell networks.The same essentiality analysis can also help forecast toxicity for other drugs.

2.4. Network comparison and similarity

Network comparison and similarity methods support node and edge prediction, molecular-network querying, and discovery of conserved or disease-specific relationships, but remain computationally demanding.

Applications: Comparing networks can uncover conserved interactions, analogous regulatory and signaling relationships, protein functions, and disease-specific changes relevant to drug design.The review names interologs, regulogs, signalogs, and phenologs as cross-network correspondences.
Limitations: Network comparison is computationally expensive and remains one of the field’s greatest challenges despite its predictive and drug-design potential.This cost is the principal scope boundary identified for the reviewed comparison methods.
Similarity measures: Similarity measures include edit, sampling, cut, similarity, and combined edit–spectral distances for quantifying relationships between networks.Edit distance counts edge changes needed to transform one network into another.
Molecular-network comparison: Molecular-network comparison often queries small subnetworks of 3 to 5 nodes against larger networks to reduce motif-search complexity.Recent methods expand beyond direct neighborhoods and compress networks into meta-nodes.

2.5. Network dynamics

The review treats molecular networks as dynamic systems whose topology, modules, robustness, and emergent behavior change over time and under perturbation, creating opportunities and modeling challenges for drug design.

Temporal dynamics: Temporal network analysis adapts connectivity, diameter, centrality, motifs, and modules to describe changing network structures and their evolution.Observation duration and time windows are crucial for detecting contacts and temporal changes.
Module dynamics: Network modules may grow, contract, merge, split, appear, or disappear, and change-point analysis identifies short periods of large modular reorganization.Alluvial diagrams visualize these structural transitions, including extreme network disintegration.
Module dynamics: Change points and topological phase transitions had not yet been assessed in disease or other therapeutically relevant abrupt events such as apoptosis.The review identifies these applications as a future drug-related research area.
Robustness and perturbation: Cellular-network robustness preserves function under perturbation, but can make disease-affected cells or parasites resistant to drug action.Robustness analysis has been used to reveal primary drug targets and network vulnerabilities.
Robustness and perturbation: Hubs, inter-modular overlaps, bridges, and small network skeletons can efficiently mediate perturbations and govern network oscillations.Viral-protein targets were also identified as major perturbators of human networks.
Cooperation and influence: Spatial games model molecular cooperation and can identify influential nodes for establishing, maintaining, or breaking cellular cooperation.The NetworGame program simulates these games on user-defined molecular networks to support future target identification.

2.6. Limitations of network-related description and analysis methods

Network-based methods remain constrained by data quality, incomplete dynamic descriptions, information loss during network construction, and the need for comparative validation. These limitations restrict confidence in where network analysis most efficiently supports drug design.

Data quality is the foremost limitation because network descriptions depend on accurate, sufficiently covered input datasets.High-quality datasets, particularly for system dynamics, are often unavailable.
Network construction can lose information when node and edge definitions reduce multidimensional biological relationships to simpler connections.Hypergraph representations may address this issue but are not yet widespread or well documented.
Network visualization still needs improved tools for three-dimensional, large-capacity, and zoom-in exploration.
Network methods require further comparative studies because molecular applications are relatively recent and their most efficient drug-design uses remain unclear.The review frames future comparison as necessary for clarifying suitable application areas.

3. The use of molecular networks in drug design

The review organizes molecular networks spanning chemical substances, protein structures, interactions, signaling, genetic and chromatin relationships, and metabolism. It emphasizes that networks may encode either physical or conceptual connections and selectively focuses on drug-development relevance.

The section surveys chemical, protein-structure, protein–protein interaction, signaling, genetic, chromatin, and metabolic networks.
The review focuses on the network features most relevant to drug development rather than detailing every study in these fields.
Network nodes and edges can represent physical relationships or conceptual associations, depending on the molecular system and analytical purpose.Examples include chemical transformations, shared binding proteins, and enzyme-mediated substrate–product relationships.

3.1. Chemical compound networks

Chemical compound networks represent molecular structure, reactions, similarity, therapy relationships, and system-wide readouts. Across these forms, network topology supports chemical-space exploration, activity and bioavailability prediction, lead optimization, target discovery, and drug repositioning.

Chemical similarity networks are especially useful for lead optimization and selection of drug candidates.
Chemical structure networks: Chemical structures can be modeled as labeled atom nodes connected by labeled covalent-bond edges, with network descriptors supporting QSAR and QSPR models.
Chemical reaction networks: 7 million compounds formed the 2012 chemical reaction network, whose hubs support synthesis planning and reveal a core–periphery organization.The core contained over 70% of the top 200 industrial chemicals.
Chemical similarity networks: Similarity-network hubs can prioritize fragment-based design, while excluding fragments near non-hit hubs can reduce chemical-space trials relative to random or cluster-center selection.This strategy depends on the assumption that molecules similar to non-hits are also non-hits.
QSAR and chemoinformatics: QSAR-related networks encode potency and landscape features, distinguishing activity cliffs from smooth regions where structural changes have different activity effects.
QSAR and chemoinformatics: Network analysis incorporating chirality can guide synthesis toward potentially high-activity derivatives, including among more than 1600 unexplored chiral inhibitor derivatives.
Bioavailability: Chemical-fragment networks identified polar and hydrophobic fragments associated with serum-albumin binding, aiding bioavailability prediction.
Network evolution and applications: Network growth around early hubs can reveal over-sampled, redundant chemical regions and redirect exploration toward less studied chemical space.

3.2. Protein structure networks

Protein structure networks model amino-acid contacts to connect protein architecture with dynamics, disease-related features, allostery, and drug binding. Their small-world and modular organization supports mechanistic analysis, but predictive use remains limited and dynamics are not fully captured.

Network representation: Protein structure networks represent amino-acid side chains as nodes and spatially close residues as edges, commonly using a 4–8.5 Å distance cutoff.
Topology and dynamics: Their small-world organization enables rapid propagation of drug-induced conformational changes across amino acids.
Topology and dynamics: Modules often correspond to protein domains, while high-centrality inter-modular bridges participate in hem binding and allosteric-change transmission.
Evolution and dynamics: Protein sectors form sparse, independently operating amino-acid networks spanning large protein regions, with co-evolving pairs often clustered in flexible regions.
Allostery: Allosteric signals may range from switch-like changes involving few residues to distributed signaling through multiple trajectories converging near inter-domain boundaries.
Allostery: Rigidity-front propagation is proposed as an allosteric mechanism in which sequential segment rigidization accelerates conformational-change transmission.
Disease proteins: Disease-related proteins tend to be longer, less designable, and structurally constrained, while disease-associated mutations often occur at locally central sites.
Drug binding sites: Elastic-network molecular dynamics predicted ligand-preferred receptor conformers and a novel allosteric binding site for larger drugs such as salmeterol.Different conformations participate in different metabolic and signaling pathways.

3.3. Protein–protein interaction networks (network proteomics)

Protein–protein interaction networks represent proteins and their physical interactions, but their probabilistic, context-specific structure and incomplete data require careful interpretation. Their topology helps identify disease-related proteins, drug targets, and interaction-specific interventions.

Protein–protein interaction networks use proteins as nodes and direct physical interactions as edges, with edge weights reflecting interaction probabilities or confidence scores.
Interactomes are species-, cell-type-, compartment-, and time-specific, so disease- or treatment-specific subnetworks may better represent drug action.
Sampling bias, missing interactions, false positives, and low conservation limit interactome robustness; confidence scores can partly address these biases.
Interactomes exhibit small-world, hub, and hierarchical modular organization, with overlapping modules often corresponding to cellular functions.
Disease-related proteins often have low clustering and form overlapping modules, while 59 core modules were affected across more than half of 54 diseases.
Drug targets generally have more neighbors than average but are usually non-hub bridges between modules, supporting controlled network perturbation and reduced side-effects.
Specific interaction modulation can provide greater disease-restoring specificity than whole-protein targeting, motivating edgetic drugs and drug-induced interactome analysis.

3.4. Signaling, microRNA and transcriptional networks

Signaling, transcriptional, and microRNA networks provide structured, dynamic representations of information flow from extracellular signals to gene regulation. Their topology and dynamics support target identification while exposing challenges in specificity and pharmacological intervention.

Signaling networks combine intertwined upstream pathways with downstream transcription-factor, DNA-binding-site, and microRNA regulatory subnetworks.
Signaling pathways transmit ligand-derived information through receptors and mediators to transcription factors, and are connected by directed physical cross-talks.
MicroRNAs regulate target-mRNA translation, and their broad transcript-level effects make them promising but pharmacologically challenging intervention points.
Signaling dynamics depends on component abundance, complex formation, crowding, and localization, and can be modeled using perturbation analysis, differential equations, fuzzy logic, or Boolean methods.
Network hubs and bridge proteins between signaling modules, including SHC, SRC, and JAK2, have been identified as recurrent drug-action targets.
Protein phosphatase targeting remains difficult because high domain homology limits selectivity, despite phosphatases shaping phosphorylation-system behavior.

3.5. Genetic interaction and chromatin networks

Genetic interaction and chromatin networks connect perturbation phenotypes to pathway mechanisms, regulatory organization, and drug prioritization. Their utility is substantial but constrained by weak conservation across model organisms.

Genetic interactions compare single-mutant phenotypes with double-mutant phenotypes, distinguishing negative interactions from unexpectedly beneficial positive interactions.
Less than 5% of synthetic lethal interactions were conserved between yeast and worm, limiting direct transfer from model organisms to human disease biology.
No single inference method performs optimally across datasets, so integrating multiple methods can improve genetic interaction network reconstruction.
Mapping genetic interactions onto physical or signaling pathways distinguishes between-pathway, within-pathway, and indirect interactions for mechanism and target analysis.
Approximately 40% of yeast synthetic lethal interactions mapped to physical pathway models, identifying 360 between-pathway and 91 within-pathway models.
Hi-C and ChIA-PET enable functionally associated human chromatin contact networks with modules and hub–hub rich-club interactions.
Genetic interaction networks can rank antifungal targets, illuminate human drug mechanisms, and assess drug combinations through effect-radius comparisons.

3.6. Metabolic networks

Metabolic networks represent metabolites and enzyme-catalyzed transformations, enabling systems-level analysis of essentiality, flux, and disease-specific vulnerabilities. Their applications include antimicrobial target selection, cancer target prediction, and compensatory therapy design.

Metabolic networks use metabolites as nodes and biochemical transformations as edges, with edges also representing the enzymes catalyzing those reactions.
Network reconstruction integrates genome sequences, enzyme databases, transcriptomes, and proteomes, but metabolic data remain incomplete and often reflect optimal growth conditions.
Flux balance, flux-variability, elementary-flux-mode, and metabolic-control analyses characterize metabolic responses at appropriate network scales.
Topology alone cannot establish enzyme essentiality because activity depends on environmental conditions, gene expression, and signaling- or interaction-mediated regulation.
Choke-point inhibition can cause lethal metabolite deficiency or toxic accumulation, making uniquely producing or consuming reactions potential infectious-disease targets.
Metabolic network analysis ranked more than a million small molecules for potential antimicrobial scaffolds after identifying common targets across E. coli and Staphylococcus aureus.
Cancer-specific metabolic analysis predicted 52 cytostatic targets, with 40% already targeted by known anticancer drugs and the remainder representing new candidates.
Flux-balance models can identify synthetic rescues, exploiting compensatory inhibition when restoring a disease-impaired reaction is pharmacologically difficult.

4. Areas of drug design: an assessment of network-related added-value

Network-related methods add value across the drug development process, from target identification through clinical trials, while supporting lead optimization and safety-related assessment.

Network methods contribute across target identification, hit finding and confirmation, lead selection and optimization, and clinical trials.The process also includes chemoinformatics, drug-efficiency optimization, ADMET studies, and optimization of interactions, side-effects, and resistance.

4.1. Drug target prioritization, identification and validation

The section presents network-based approaches for prioritizing and identifying drug targets, including central-hit and network-influence strategies, edge targeting, polypharmacology, and allo-network action.

Two strategies of network-based drug targeting: Central-hit targeting attacks central nodes or edges in infectious-agent and malignant-cell networks, whereas network influence seeks to reconfigure diseased networks.The latter requires understanding network dynamics in healthy and diseased states.
Two strategies of network-based drug targeting: Flexible networks are targeted at central nodes or edges, while rigid, multi-modular networks are more effectively influenced through neighboring or inter-cluster nodes.Central nodes transmit attacks efficiently in flexible systems; flexible connectors can dissipate perturbations in rigid systems.
Network target prioritization: Influence-cores may regulate many diseases beyond connection-cores or network peripheries, with 59 dysregulated interactome modules enriched in drug targets across at least half of 54 diseases.The module count and disease coverage are reported from an interactome analysis.
Edgetic drugs: Edgetic drugs selectively perturb protein–protein or other network interactions, potentially addressing nonenzymatic proteins and separating disease-related functions of multifunctional proteins.The paper highlights mTOR as an example where one protein participates in complexes with different functions.
Edgetic drugs: A yeast assay tested 80 diverse small molecules and found that FK506 specifically inhibited the interaction between aspartate kinase and Fpr1.The study illustrates experimental identification of small-molecule-induced interactome changes.
Drug target networks: Binding-site and drug-target network similarities support repositioning and novel target prediction, while many drugs act on proteins near disease-associated proteins rather than on those proteins directly.Network comparisons can be more pharmacologically informative than sequence or structure comparisons alone.
Network polypharmacology: Polypharmacology combines modulation of multiple targets, and more than 20% of approved drugs are multi-target drugs.Partial, low-affinity attacks across several network sites can outperform complete inhibition of one node in reported E. coli or yeast regulatory-network studies.

4.2. Hit finding, expansion and ranking

Network-based hit discovery combines structural, chemical, efficacy, toxicity, and resistance analyses to prioritize tractable and safer drug candidates. These methods extend from binding-site prediction and synthesis planning to side-effect forecasting and resistance-aware combination design.

Hit finding: Binding-site discovery uses protein structure networks from the bottom up and binding-site similarity networks from the top down.Protein pockets and transport pathways provide structural features for in silico hit prediction.
Hit finding: High-centrality protein-structure segments and conserved protein sectors can identify catalytic and allosteric ligand-binding sites.These approaches connect network centrality and evolutionary conservation with ligand-binding-site prediction.
Hit confirmation and lead selection: Chemical-reaction-network centrality predicts chemical tractability, while simulated annealing can optimize synthetic pathways for selected hits.Core and hub positions are associated with feasible synthesis and support lead-selection decisions.
Efficacy and personalized medicine: Network-based efficacy profiling can integrate pharmacogenomics, signaling, metabolome, and medical-record data, but complex efficacy methodologies remain undeveloped.Network models may also help identify effective dose and schedule regions.
Toxicity assessment: Network methods assess toxicity through perturbation, robustness, centrality, and signed toxicity-promoting or toxicity-reducing interactions.Inter-modular bridges may be preferred over hubs for some interventions, while drug-regulated-network centrality correlates with toxicity.
Network pharmacovigilance: Side-effect prediction combines side-effect, biological-process, literature, disease-gene, and interactome data because adverse effects can involve single or multiple targets.Two thirds of side-effect similarities were related to shared targets, whereas 5.8% involved proteins close in the human interactome.
Side-effect reduction: Avoiding hubs and high-centrality nodes limits network perturbation in network-influence strategies, while influential disease-state nodes may offer lower-side-effect targets.The proposed boundary is disease-specific perturbation rather than indiscriminate network disruption.
Resistance-aware design: Resistance can arise through alternative or counteracting pathways, motivating network-weighted co-targeting and antagonistic combinations based on synthetic rescues.The paper contrasts these options with highly synergistic combinations that may select resistance faster.

5. Four examples of network description and analysis in drug design

Network description and analysis are applied across infectious disease, cancer, and metabolic disease to identify disease-related targets and model context-specific intervention. The examples show that integrating molecular layers and network dynamics can refine target prediction, reveal metabolic candidates, and guide combination therapies.

Infectious disease: Combining viral-host interactomes with siRNA, transcriptome, microRNA, and toxicity data can extend antiviral target prediction.Network information also identified formerly validated influenza A targets and predicted novel candidates.
Cancer: Cancer-specific interactomes place mutated proteins at hubs, rich clubs, bridges, bottlenecks, and other central positions.These positions are used in network-based cancer target prioritization.
Cancer: mRNA expression alone is generally insufficient for target-efficiency prediction, whereas microRNA, proteomic, domain, ontology, mutation, and prognosis data refine predictions.Random-walk interactome methods identify subnetworks around proteins with altered levels and evaluate associated mRNA dysregulation.
Cancer metabolism: Cancer-specific metabolic-network analysis predicted 52 cytostatic drug targets, including 40% targeted by known anticancer drugs and additional new candidates.The analysis reflects cancer metabolism adapted to proliferative needs under predominantly anaerobic conditions.
Cancer signaling: Cancer signaling therapies aim to rewire altered networks, whose greater complexity has been associated with shorter survival.Cancer-mutated proteins are often hubs in human signaling networks.
Cancer signaling: mTOR illustrates the difficulty of node targeting when a mutated, hyperactive protein participates in multiple signaling pathways.Despite its expected therapeutic promise, mTOR-directed drugs showed poor clinical-trial results.
Cancer regulation: Cancer-specific microRNA networks show hub dysregulation and more disjoined subnetworks than normal-tissue networks.Combined mRNA and microRNA data can infer cancer-specific regulatory networks.
Cancer dynamics and therapy: Network dynamics can prioritize cancer targets by modeling perturbation dissipation, while personalized mutation, signalome, and metabolome profiles may refine targeting.Cancer robustness and redundant pathways motivate multi-target and combination therapies.

6. Conclusions and perspectives

The review positions network description and analysis as a complement to creativity and background knowledge for discovering and optimizing drugs. It proposes distinct targeting strategies, alternating exploration with optimization, and systems-level hallmarks to improve drug development.

Targeting strategies: Central hit strategy targets key nodes in flexible pathogen or cancer networks, whereas network influence strategy reconfigures rigid disease networks toward normal dynamics.The two strategies are framed as general approaches for attacking flexible versus rigid systems.
Promises and optimization: Network analysis should be combined with human creativity and background knowledge to identify surprisingly novel drugs.Unbiased methods may predict novel targets but can miss true surprises.
Protocol: Network-aided drug development alternates exploration, which suppresses background knowledge, with optimization, which rigorously ranks options using that knowledge.The protocol contrasts exploratory playfulness and ambiguity tolerance with optimization's rule-based evaluation.
Limitations: The field still lacks rigorous comparisons, benchmarks, gold standards, and assessment tools for many network-science methods in drug design.The review characterizes network science as a relatively novel area of biology, particularly for drug design.
Drug-quality trends: The review highlights edgetic, multi-target, and allo-network drugs, alongside patient-specific biological data, human ADME and toxicity models, and side-effect networks.These trends are presented as ways to achieve systems-level hallmarks of drug quality and improve development efficiency.
Conclusion: Network methods are proposed to uncover novel targets and increase development efficiency when applied systematically and carefully.The proposed paradigm extends network analysis to molecular structure, dynamics, proteins, and binding sites.