The Hidden Science of Drug Purification After AI Discovery
AI is supercharging drug discovery, but manufacturing can't keep pace. Experts reveal the critical purification challenge between lab and pharmacy shelf.
By Adam Makins · 
By Adam Makins · Artificial intelligence is flooding drug discovery pipelines with unprecedented numbers of promising candidates. Researchers can now design potential miracle molecules faster than ever before. But there’s a catch that nobody talks about at the cocktail parties where biotech breakthroughs get celebrated: we’ve become terrifyingly good at inventing drugs we can’t actually make.
The gap between laboratory success and commercial manufacturing has become the bottleneck of modern medicine. A molecule that works perfectly in a flask at small scale doesn’t necessarily work in a 10,000-liter bioreactor. The biology shifts. The engineering shifts. Everything becomes non-linear, and suddenly purification becomes the hidden battlefield where drug pipelines either succeed or fail.
The Scale Problem Nobody Expected
This is the subject of a compelling New Scientist CoLab podcast featuring experts from Cytiva, a global life sciences leader. The conversation cuts straight to the heart of an industry-wide crisis: as AI supercharges our ability to discover drugs, manufacturing capacity struggles to keep up.
Taking a drug from theoretical design to the pharmacy shelf involves far more than just having a good molecule. There’s the tedious work of understanding exactly what you’re purifying, then figuring out how to separate it from everything else at commercial scale. At small scale, this is manageable. At industrial scale, it becomes a completely different animal. The challenges of keeping pace aren’t just technical; they’re fundamentally about time. While AI can screen millions of compounds in weeks, scaling production still takes years.
Why does small scale success not translate to industrial scale? The physics change. Mixing patterns differ. Heat distribution varies. What worked in a lab flask might create entirely new problems when you’re dealing with thousands of liters of biological material. The molecules themselves can behave differently under stress.
When Purification Goes Wrong
The stakes here are impossibly high. When purification fails or falls short, the consequences aren’t abstract. Real patients waiting for treatments face delays. Supply chains fracture. Clinical trials stall. In some cases, contamination or ineffective purification means a potentially life-saving drug never reaches the people who need it.
The podcast explores the human impact of these failures, reminding us that pharmaceutical manufacturing isn’t just about chemistry and engineering. It’s about real people whose lives depend on getting these processes right.
The expanding drug pipeline adds another layer of complexity. AI isn’t slowing down. If anything, it’s accelerating. More candidates mean more demands on purification infrastructure that was never designed to handle this volume. The industry is essentially running a race against its own success.
The Invisible Science
Purification becomes progressively more difficult later in the process, partly because by the time you’re trying to purify something at scale, you understand less about it than you did in discovery. The simpler your molecule, the easier purification tends to be. But complexity increases demands exponentially.
Advancing drug discovery feels amazing when you’re celebrating a breakthrough. But someone still has to figure out how to make a billion doses of it safely. That someone is usually working in obscurity, wrestling with problems that didn’t exist before AI started generating candidates at such a furious pace.
The conversation between experts in this podcast reveals something crucial: the innovation pipeline is only as strong as its manufacturing bottleneck. Right now, that bottleneck is looking increasingly fragile. We’ve optimized discovery but not production. We’ve glorified the creation of new molecules but neglected the unglamorous science of actually making them.
The question facing the pharmaceutical industry isn’t just whether we can design better drugs with AI. It’s whether we can manufacture them. Until manufacturing catches up, all those brilliant discoveries remain theoretical achievements rather than actual medicines.
Source material from New Scientist CoLab podcast, hosted by Justin Mullins
If AI can design a drug in days but manufacturing takes years, who really controls the future of medicine?