Network Targets: How Herbal Formulas Work on Multiple Pathways Simultaneously

Imagine you are trying to understand a city by studying a single intersection. You could learn a great deal about that corner — the traffic patterns, the pedestrian flow, the timing of the lights. But you would not understand the city. Cities are not intersections. They are networks of interdependencies, where a road closure three blocks away ripples through routes you never anticipated.

Disease is more like a city than an intersection. And for the better part of a century, drug development has been studying intersections.

The Vocabulary Shift

The term "network pharmacology" was introduced by Andrew Hopkins in 2007, in a brief but consequential paper in Nature Biotechnology. The concept was simple and radical at the same time: rather than asking how a drug affects a single target, ask how it affects a biological network. Rather than optimizing for potency at one site, optimize for modulation across many.

Hopkins was building on a body of evidence that had been accumulating for years. Drug-target network analyses were revealing that most drugs — including many designed for single targets — actually bind to multiple proteins. What had been called "side effects" were, in many cases, effects on unintended targets in the same network. Polypharmacology, the property of a drug acting on multiple targets, was not an anomaly. It was the norm.

The implication was uncomfortable for the reductionist model but clarifying for everything else: if most drugs are already multi-target, and if many diseases involve dysregulation across multiple pathways simultaneously, then perhaps the goal should never have been to find a single perfect key for a single perfect lock. Perhaps the goal should be to modulate a network.

What a Network Target Actually Is

A network target is not simply multiple targets. The distinction matters. Multiple targets means a drug binds several proteins. A network target means the relevant unit of therapeutic action is a pattern of interactions within a biological network — a set of nodes and edges whose collective modulation produces the desired clinical effect.

To build that picture, researchers construct several overlapping maps. A protein-protein interaction (PPI) network connects proteins based on known binding and functional relationships. A drug-target network links compounds to the proteins they interact with. A disease-gene network maps the genetic underpinnings of a condition. When these networks are layered, the overlapping regions — where a drug's targets intersect with a disease's genetic drivers — reveal the network target.

This is not theoretical. For complex diseases like cancer, cardiovascular disease, and neurodegeneration, researchers have used these overlapping networks to identify key hub proteins — highly connected nodes whose modulation affects large portions of the disease network. A single compound that hits one such hub can have cascading effects across dozens of downstream pathways. A formula that hits several hubs simultaneously can reshape the network in ways no single compound could.

Why Herbal Formulas Are Built for This

A classical TCM formula is not a mixture of random compounds. It is a structured polychemical system, refined over centuries of clinical observation, in which individual herbs play defined functional roles. The jun-chen-zuo-shi framework — roughly translated as sovereign, minister, assistant, and envoy — is a compositional logic that distinguishes primary therapeutic agents from those that support, modulate, or direct them.

What classical practitioners could not see at the molecular level, network pharmacology is now beginning to reveal. When researchers apply network analysis to well-characterized formulas, they consistently find that the compounds act on multiple biological targets simultaneously — often across several interconnected pathways — and that the interactions between compounds are not merely additive. They are synergistic: the combined effect on the network exceeds what the components would produce individually.

Take quercetin and berberine — two compounds found across multiple classical formulas. Both appear in network analyses of colorectal cancer, cardiovascular disease, and neurodegeneration. Both modulate the PI3K/Akt signaling pathway, one of the most consequential hubs in cancer biology. But they enter that network at different nodes and via different mechanisms, meaning a formula containing both produces a richer, more distributed intervention than either alone. Network pharmacology studies of formulas like Sijunzi decoction and Zuojinwan have mapped exactly this kind of multi-node engagement across the PI3K/Akt/mTOR axis in colorectal cancer cells — identifying not just that the formulas work, but structurally how they work.

The Computational Infrastructure

None of this analysis would be possible without a substantial infrastructure of databases and tools that has been built, largely over the last two decades, to support network pharmacology research.

On the TCM side, databases like TCMSP (Traditional Chinese Medicine Systems Pharmacology) and HERB catalogue the chemical constituents of thousands of herbs, their known bioactivities, and their absorption and pharmacokinetic properties. On the disease side, GeneCards, DisGeNET, and OMIM map the genetic and molecular architecture of human conditions. STRING and BioGRID maintain protein-protein interaction networks covering tens of thousands of proteins. Cytoscape provides the visualization and analysis layer that lets researchers actually see the networks they are building.

A typical network pharmacology study of a TCM formula follows a defined workflow: identify the active compounds in the formula and filter them by bioavailability parameters; query their known targets; cross-reference those targets with the disease gene set; construct the compound-target-disease network; identify hub nodes and key pathways; validate computationally via molecular docking; validate experimentally in cell or animal models. The field has moved quickly enough that there are now published guidelines for this workflow, and the number of studies using it has grown dramatically since AI tools began accelerating the target prediction and network construction steps.

Not all analytical approaches within this workflow perform equally well, however. The most commonly used methods — measuring how much a formula's targets overlap with disease proteins, or how close they sit within a protein-protein interaction network — turn out to have meaningful limitations when tested rigorously against known outcomes. A 2025 systematic evaluation by Lee et al. compared these approaches head-to-head and found that modeling how a formula's targets propagate their influence across a network that integrates both protein interactions and biological functions — a multiscale interactome — consistently outperformed simpler proximity measures. The method is more computationally demanding, but it captures something the simpler approaches miss: that therapeutic effects arise not just from where a drug's targets sit in a network, but from how their influence diffuses through layers of biological organization.

The Limits of the Map

Network pharmacology is a map, not the territory. Like all maps, it simplifies. And like all maps, it depends heavily on the quality of the underlying survey data.

That dependency turns out to be a significant practical problem. When Lee et al. systematically compared five major TCM network pharmacology databases — querying the same herbs across all five — they found that while ingredient information was reasonably consistent, target information for identical compounds diverged substantially between databases. None of the database pairs achieved 90% recall consistency for compound-target associations. In some cases, one database listed over 1,600 unique targets for compounds where another listed far fewer. Which targets a researcher finds depends, in no small part, on which database they happen to use. The mechanism of action they report is partly a function of their data source.

The networks are also largely static, capturing average interaction states rather than the dynamic, context-dependent, tissue-specific behavior of living biological systems. And the passage from computational prediction to clinical validation remains long and difficult.

There is also a deeper conceptual challenge. Network pharmacology was developed primarily within a Western disease taxonomy — conditions defined by molecular and pathological criteria. TCM operates with a different nosology: syndromes defined by patterns of symptoms, constitution, and functional state. Translating between these frameworks is not a solved problem. A formula indicated for a particular TCM syndrome may map onto several different biomedical conditions, or none exactly. The computational infrastructure is improving, but the conceptual bridge is still being built.

A New Kind of Question

What network pharmacology has changed, fundamentally, is the kind of question researchers can ask about botanical medicines.

The old question was reductionist by necessity: which single compound in this herb is responsible for its effect? That question produced some real discoveries — artemisinin from Artemisia annua, berberine from Coptis chinensis, paclitaxel from yew bark — but it also produced a methodological bias. Formulas were treated as impure versions of the compounds they contained. The complexity was seen as a problem to be simplified away.

The new question is systemic: how does this formula interact with this disease network? That question is harder to answer, but it is closer to the right question. It takes the complexity of the formula seriously, as a feature rather than a flaw. It asks not which molecule does the work, but how the system as a whole produces its effects — and which nodes, pathways, and interactions are essential to that production.

That is the question network pharmacology was built to answer. And it is the question that, increasingly, the field has the tools to address.

"Network targets emphasise understanding the broader biological mechanisms — not just the drug's effect on one or multiple targets, but how the system as a whole is modulated."
— Joshi et al., Naunyn-Schmiedeberg's Archives of Pharmacology, 2025