AI Drug Discovery: When Data Becomes the Real Recipe

Something strange happens in pharma: researchers have access to mountains of data — tested molecules, clinical results, protein structures — but often don't know how to make them talk to each other. It's like owning a giant library where every book is in a different language and nobody has a dictionary. AI, in this sense, plays the role of the smart translator.

According to Pradeep Bandaru, who leads AI workflows at Sanofi, the real game-changer isn't AI itself — that's commoditized by now — but the quality and integration of your data. When you can connect the dots between failed trials, molecular structures, and biological breakthroughs, suddenly the computer starts spotting patterns no human would see in a lifetime. Not because it's magic, but because it's working at a scale and speed impossible for the human brain.

Shawn Rosemarin and his team understand the practical challenge is different from the lab dream. It's not just 'using AI,' but building pipelines that let systems analyze messy, incomplete, and often contradictory data. Sanofi is betting hard on this — not to replace chemists, but to give them a sharper spear. A researcher with AI on their side isn't a worse scientist; they're a scientist who can test ten times more hypotheses in the same timeframe.

Here's the interesting bit: the real bottleneck isn't computing power anymore — that exists and keeps getting cheaper. It's the ability to organize, clean, and standardize data so AI can actually use it without losing its mind. It's the unglamorous work nobody writes papers about, but it's what separates a prototype from medicine that actually works.

That's where Rosemarin's 'customer engineering' philosophy comes in: instead of waiting for data to become perfect (it won't), you build robust systems that learn to work with real-world data, messy as it is. It's a mindset shift — away from the obsession with data purity and toward accepting that disorder is our natural habitat.