- Oops!Something went wrong.Please try again later.
The field of drug discovery has been supercharged by the capabilities of AI, which several companies have applied in various ways to turn an enormous practical problem into a tractable information problem. The latest to do so is Google parent company Alphabet, which has established Isomorphic Labs, under DeepMind head Demis Hassabis, to take its shot at the promising new field.
Very little was revealed about the company in its debut blog post and a very general accompanying FAQ. The aim of the company is to "build a computational platform to understand biological systems from first principles to discover new ways to treat disease."
There are, of course, a few assumptions baked into that founding statement, most prominently that it's possible to computationally simulate biological systems in a matter conducive to drug discovery.
Several large companies have been formed and funded with hundreds of millions of dollars to pursue very similar goals over the last five years or so, and there has been no visible revolution or famous AI-discovered wonder drug for a previously untreatable disease. The reasons why are beyond the scope of this article (and will no doubt be engaged with by Isomorphic Labs in the near future) but it is clear these AI systems are not miracle factories, just parts in a long and complex process that still involves a great deal of time, money and test tubes.
Hassabis is no fool, and although he describes biology rather optimistically as "an information processing system, albeit an extraordinarily complex and dynamic one," (I can sense the readers in this field scrolling down to the comments now) he tempers that shortly afterwards:
Biology is likely far too complex and messy to ever be encapsulated as a simple set of neat mathematical equations. But just as mathematics turned out to be the right description language for physics, biology may turn out to be the perfect type of regime for the application of AI.
The idea that information systems and biological systems may have a common structure is the inspiration for the name, Isomorphic Systems; isomorphic means alike in shape but having a different origin.
His reasoning for this is no doubt partly from the effectiveness of DeepMind's AlphaFold, an AI-powered protein folding system that blew biologists' socks off last year and helped create a new normal in a very complex field.
DeepMind's learning systems have shown a particular affinity for generality or knowledge transfer — that is, having a structure that is capable of being repurposed for very different tasks. And if, as AlphaFold's success suggests, biological systems are a good match for this kind of simulation and analysis, Hassabis's assessment of the company's wider capabilities may prove true.
If so, it won't be for a while. Even with the running start provided by DeepMind's AI research (which will remain separate but may be shared), Isomorphic is essentially starting from scratch on this problem. It's hiring up a "world-class multidisciplinary team" and perhaps in a year or two we may see the first inklings of results issuing from the company's ambitions.