How systems biology and network medicine may help solve the biggest challenge in drug discovery: human biology.
When I was studying pharmacokinetics between 2005 and 2010, the field was hitting its stride. Lipinski’s Rule of Five had already reshaped how medicinal chemists thought about drug-likeness. Population PK modeling and early PBPK tools like Simcyp were just maturing; we were learning to use them in real time. The goal was clear, integrate all available data to get the most optimal molecule to the clinic, and stop losing drugs to poor pharmacokinetics.
It worked. ADME-related failures dropped dramatically. The industry identified the bottleneck, built the tools, and solved it.
But the overall failure rate barely moved.
Because we had solved the wrong problem first. And in drug discovery, the wrong problem is chemistry. The right problem, the one that still drives a 90% clinical failure rate, is target validation and disease biology.
Why AI Drug Discovery Is Winning in Phase I But Stalling in Phase II
History is repeating itself.
AI has done for chemistry what computational PK did for pharmacokinetics. Generative models now explore chemical space at a scale no human team could match. Binding predictions are faster and more accurate. Timelines from target to pre-clinical-candidate are compressing. According to a first-of-its-kind analysis by BCG published in Drug Discovery Today, Phase I success rates for AI-discovered compounds are running at 80–90% — far above the historical industry average of 40–65%.
Impressive.
But across the field, Phase II success rates remain at roughly 40%, in line with historical averages. Basically, unchanged. However, to be fair, there are early signals of genuine clinical progress. Insilico Medicine published positive Phase IIa results for rentosertib in Nature Medicine in June 2025 (the first clinical proof-of-concept for an end-to-end AI-discovered drug). A meaningful milestone.
Because Phase II was never a chemistry problem.
The Real Reason Drugs Fail: Target Validation, Not Chemistry
Drugs die on efficacy.
Not toxicology. Not ADME. Efficacy.
We pick targets that are biologically real but causally ineffective. The molecule does exactly what it was designed to do but the patient doesn’t get better. That gap existed long before anyone said the word “AI.” Faster chemistry doesn’t close it.
The question that deserves more attention is a harder one: why does modulating this target, in this disease, in a living human being, actually change anything?
Systems Biology and Network Medicine: The Next Frontier
We’ve been treating biology like a list.
One target. One pathway. One intervention. Many R&D teams still operate within this paradigm, and this model has produced real medicines. But some teams are already asking a harder question. The emergence of network biology, multi-omics integration, and causal AI is beginning to reframe how we think about disease; not as a broken gene or a dysregulated pathway, but as a system that has shifted into an unhealthy state. The goal isn’t to hit a target. It’s to understand which intervention moves the system back.
Biology doesn’t work like a list. It’s a network, layered, dynamic, full of feedback loops and compensatory mechanisms that only reveal themselves in the context of a whole system. A target that looks causal in isolation can be irrelevant (or worse, counterproductive) when the network reorganizes around it.
This is what systems biology has been telling us for two decades. The tools to act on it at scale are finally here.
AI and Network Medicine: Where the Next Breakthrough Will Come From
The next frontier isn’t better generative chemistry.
It’s AI applied to the network itself. Understanding how disease emerges from the interaction of hundreds of nodes (genes, proteins, pathways, cell states, and microenvironment signals) and identifying the true leverage points.
The places where a small, precise intervention can propagate meaningful change throughout the whole system. Not the loudest node. Not the most obvious pathway. The right one.
Recent work is pointing exactly in this direction. A 2025 Clinical Pharmacology & Therapeutics paper built a network medicine model integrating brain-specific multi-omics data to prioritize drug targets for ALS, identifying 105 candidate genes and testing them against real-world Electronic Health Records (EHR) survival outcomes. The signal held up: chemokine receptors and neuropeptide Y emerged as promising targets, and effect sizes replicated in myasthenia gravis and Parkinson’s.
A second example makes the same case in oncology. A Communications Chemistry study combined single-cell transcriptomics, CRISPR screens, and protein interaction networks to prioritize targets in renal cell carcinoma, then validated them experimentally. ENO2 inhibition showed the strongest anti-tumor effect, followed by LRRK2 (a repurposing candidate already in Phase III for Parkinson’s).
Different diseases, different data types, same move: target validation built from real biological signal, not just chemical druggability. The early signals are genuine.
The question now is who can build and industrialize these systems-level capabilities fast enough, and whether capital allocators and the broader industry know how to recognize opportunities to fund them when they arrive.
If you work in biotech venture, pharma BD, or drug discovery strategy, this is the question worth following. I write about where science meets strategy, and what it means for the future of drug development.
I’ve spent my career on both sides of this equation, optimizing molecules to get them to the clinic, and now evaluating assets worth betting on.
What excites me right now isn’t AI for the fastest chemistry. It’s understanding why a target matters inside a living system. How it influences disease. How getting it right can transform someone’s life.
With the excitement of the World Cup, I would say:
Match one: chemistry. ⚽ 1–0.
Match two: biology.
What are you watching in this space? Drop it in the comments.