AI's bold mission to 'solve all diseases'

In the relentless pursuit of medical breakthroughs, one company stands at the intersection of artificial intelligence and drug discovery with a vision that borders on the fantastical: solving all diseases. That company is Isomorphic Labs, and its Chief AI Officer Max Jaderberg recently shared insights into how they're reimagining the entire drug discovery process using AI capabilities that were previously unimaginable.

The scale of ambition is breathtaking

Isomorphic Labs isn't focused on developing therapeutics for a specific indication or target. Instead, they're building what Jaderberg describes as a "very general drug design engine" – something that can be applied across any disease area or molecular modality. This approach represents a fundamental shift from traditional pharma, which typically builds expertise around specific disease categories or molecule classes.

The scale of the challenge is immense:

• The potential chemical space for drug-like molecules is estimated at 10^60 – even if we reduce that by 20 orders of magnitude, we're still looking at 10^40 possible molecules to explore
• Traditional predictive models, even if they could screen a billion molecules, would barely scratch the surface of this vast chemical space
• Creating truly transformative drug design capabilities requires "half a dozen AlphaFold-level breakthroughs" across different domains of biology and chemistry

Perhaps most fascinating is Jaderberg's description of the algorithmic approach. He draws direct parallels to his previous work on game-playing AI systems like AlphaGo and Capture the Flag, explaining that drug discovery requires both powerful predictive models (understanding the game) and generative agents that can navigate the vast possibility space (playing the game expertly).

AlphaFold 3: Visualization meets imagination

The most significant breakthrough so far is AlphaFold 3, which has expanded from predicting protein structures to modeling how any molecules interact with each other in three-dimensional space. This capability transforms drug design from a partially blind process to one where designers can immediately visualize how molecular changes affect interactions with target proteins.

What makes this particularly powerful is the end of what Jaderberg calls "local models" – narrow AI systems trained on specific targets or molecule classes. AlphaFold 3