Systems pharmacology
Classical pharmacology finds a receptor, designs a molecule that binds it, and measures what happens — clean, and often not enough. Biology doesn't respond to a drug by flipping one switch; it responds through cascades of interacting proteins, feedback loops, and compensatory pathways that a single-target experiment can't see. That's a real part of why drugs that work cleanly in a test tube often fail in patients: the target genuinely binds, but the network around it responds in ways the isolated experiment never predicted.
Systems pharmacology takes the network seriously, combining three things at once: mechanistic biology (the actual signalling pathways and gene regulatory networks a drug touches), mathematics (ordinary differential equations and network models that simulate how those pathways behave over time), and pharmacokinetics (drug concentration, integrated with the network model so it captures how the biological system evolves over days or weeks of treatment, not just at one instant). The result is a computational model of how a disease's biology responds to treatment — including feedback loops, resistance mechanisms, and combination effects — built to inform trial design, dose selection, and which patients are likely to respond at all.
In industry, this discipline is usually called QSP — quantitative systems pharmacology — and it's the frontier that pharmacometrics is actively expanding into as drug development moves further into combination therapies and biologics.
Where it sits in development
Systems pharmacology sits earlier than most of the PK/PD work in this pillar — primarily discovery and translational science, though it's increasingly reaching into later clinical stages. In discovery, it's used for network-level target identification: which node in a disease network, if perturbed, actually shifts the system toward health. In translational work, it supports animal-to-human dose prediction. In Phase II, it can guide which measurable biomarker actually predicts clinical benefit, and in Phase III, which patient subpopulation is likeliest to respond, based on their baseline biology. Regulatory use is still early — the FDA has been exploring systems-pharmacology models specifically in oncology, but this isn't yet a routine part of a submission the way a PopPK report is.
What's actually settled, and what isn't
This is younger, as a discipline, than any other subfield in this pillar. The underlying mathematical frameworks and specific well-characterised signalling pathways (MAPK, PI3K/AKT/mTOR, and a handful of others) are established; a genuinely comprehensive, validated model of a whole disease network isn't. What's open:
- Predicting drug combination effects from first principles. Why two drugs that each work alone sometimes fail, or become toxic, together is something a systems model should in principle be able to predict — in practice, oncology combination regimens are still selected largely by empirical trial and error.
- Modelling acquired resistance — how cancer cells, bacteria, or viruses evolve under drug pressure — is only partially captured by current models.
- Multi-scale integration, connecting molecular-level biology through cellular and tissue behaviour to a clinical outcome in one validated framework, doesn't really exist yet for any disease. Some people call the eventual patient-specific version of this a "digital twin"; it's an aspiration more than a working method right now.
- Immuno-oncology specifically is a hard open problem: the tumour microenvironment is an ecosystem of cancer cells, immune cells, stromal cells, and vasculature, and only a minority of patients respond to checkpoint inhibitors — a systems model that reliably predicted who would respond, and why, would be a genuinely large result.