Forget the model wars for now. Meanwhile, there is an AI battle taking place regarding new drug development. Google’s subsidiary, DeepMind, has been developing protein folding for several years now and is using its spinoff, Isomorphic Labs, to create drugs through this technology using $2.1 billion worth of investments. On the other hand, OpenAI has chosen another approach and has acquired Orchestra Bio in order to deal with the process of conducting all the trials needed for the development of a drug.
Also read: Workday’s Gerrit Kazmaier on why AI won’t take the human out of work
On Google’s end, it starts with the science. AlphaFold solved the problem of protein structure prediction, and the spin-off Isomorphic Labs is taking that breakthrough and turning it into drugs. In May, it raised $2.1 billion from Thrive Capital, with Alphabet’s GV and CapitalG, MGX and Temasek joining in. The company already has relationships with Eli Lilly and Novartis. In February, it unveiled IsoDDE, which it claims outperforms AlphaFold 3 in terms of biomolecular structure prediction and drug discovery.
Meanwhile, OpenAI is coming at it from the other way around. GPT-Rosalind analyzes biological and chemical data, while Rosalind Workbench provides assistance with analysis and experiment planning. The Orchestra module adds agents monitoring the entire drug program. For OpenAI, the example would be a failed manufacturing run that delays the clinical trial process, while the software warns about the domino effect and analyzes the different scenarios. The researchers and program managers make the decisions themselves. Partners of OpenAI include Novo Nordisk and Moderna.
Also read: Onimusha: Way of the Sword review: Sharp combat, uneven pacing
But consider this. It takes 10 to 15 years to develop a drug, leaving plenty of space for the errors in the handoffs. No one ever talks about the missed handoff, but this is where the months vanish.
Isomorphic’s science is outstanding, but the only arbiter that counts is the clinic. Isomorphic now plans its first clinical trials for late 2026, rather than its original plan of late 2025 made by Demis Hassabis. Predicting protein binding to a molecule and making a drug that does the job are two different things altogether.
OpenAI faces a similar gap. It has not shown a breakthrough equivalent to that made by AlphaFold in the structural prediction space. The details of the Orchestra agreement remain undisclosed. So the purchase of a workflow product is a safer move than it seems at first sight. Saved weeks of work are much more easily validated than a cured patient, and I suspect that OpenAI realizes it.
Head to head race, but maybe not. Regardless of what Isomorphic designs, it must pass through trials, production and regulation. Whoever does this, sits right in the middle of things, and that is where OpenAI wants to be.
For now the two may become complementary with one supplying molecules and another the pipeline for those molecules. The conflict arises much later down the road once both want to become the default solution pharma uses to start their day. Pharma companies tend to be conservative buyers, and you can bet this battle will take a while and cost a lot of money.
There is also a trust issue which is yet to be resolved. If the agent gives recommendations on how to address a delayed batch, who is responsible to a regulator if the recommendation fails? OpenAI insists on the man-in-the-loop approach, and auditors will demand to see it in action.
A substantial portion of drug R&D globally depends on India, from the contract research organizations to the global capability centers of Hyderabad, Bengaluru, and Pune. Coordination is very much part of their activities, making the use of program tracking agents both a danger and an opportunity. There will be entities that will be squeezed while some will discover a new product line to sell alongside the tools.
The cost-efficient Indian drug manufacturers have the most to benefit in case of the capability of the software to detect any delay before it becomes too late. This group has everything to lose in case the most effective tools remain exclusive because of enterprise contracts aimed at the likes of Novo Nordisk.
Also read: I tested ColorOS 17 on OnePlus 15 for weeks: Here is what makes the new software stand out