There was a fascinating article in the IPKat blog about the AI programme AlphaFold recently. So, what is this all about?
AlphaFold
AlphaFold is a machine-learning model for predicting the 3D structure of proteins. The most recent version, AlphaFold3, is significantly more advanced than earlier iterations - AlphaFold3 can apparently 'predict the structures of biological molecules other than proteins, including nucleic acids and lipids, and protein-drug interactions'.
Google 'believes that AlphaFold3 will place it at the forefront of AI-assisted drug discovery'.
Not for kids
This is, of course, serious stuff! You may like this 'science boff puts humanities-type in his place' line from the IPKat blog:
'Whilst large language models such as ChatGPT can write poems and make pretty pictures, AlphaFold has the potential to dramatically impact the life-and-death world of drug discovery.'
Ground-breaking
The blog describes this as 'truly ground-breaking' science, as evidenced by the award of the Nobel Prize for Chemistry. Google invested heavily in the development of AlphaFold and now wants to see a return on its investment.
We're told that, whereas early iterations of AlphaFold were open-source, it is now commercial, with AlphaFold3 being intended as a commercial product, and Isomorphic Labs (a Google Deep-Mind offshoot) leading the way.
Some intellectual property (IP)
- Trade secrets
The article says that 'trade secrets were never the IP strategy for protecting AlphaFold.' No surprise there - it would only be a matter of time before coders were able to 'reverse engineer DeepMind's approach'. Something that appeared to happen frequently after the release of ChatGPT with other models like Anthropic's Claude, Google DeepMind's Gemini and Meta's LLaMA.
- Patents
IPKat tells us that DeepMind has filed a number of patents 'covering different machine learning approaches to predicting protein structure, including many aspects of the AlphaFold architecture'.
One patent relates to the use of 'multiple sequence alignment (MSA)'. Another covers 'transformer-like mechanisms for predicting protein structure'.
- Free-for-use
AlphaFold and AlphaFold2 were provided free of charge to academics and industry. But access to AlphaFold3 is more restricted – it can only be run via Google DeepMind's API, the AlphaFold Server, and this is only available for non-commercial use. Users may not use the 'AlphaFold Server in its outputs for commercial activities or to train machine learning models similar to AlphaFold'.
- Commercialisation
The 'restrictions represent the clear intention of Google DeepMind to commercialise the model via its spin-out Isomorphic labs'.
The article goes on to suggest that Isomorphic already has drug discovery collaboration with big companies like Novartis and that such collaborations probably 'include a commercial licence for the use of AlphaFold3'.
What about academia?
The academic world is not particularly enamoured of this new commercial approach to AlphaFold3. Many say that it is in stark contrast to what happened with AlphaFold2, where the entire underlying code was made available so that researchers could test it for themselves.
Pseudocode
With AlphaFold3, however, the code has not been published. There's simply a 'pseudocode' which, as we understand it, is no more than a description of the code.
An academic outcry
The pseudocode has got scientists riled, with more than a thousand of them complaining that the commercial approach has 'compromised peer-view' and made 'the broad claims of the paper impossible to test'.
Back to IP
There is more IP in the article:
Enforcement?
The article suggests that the release of the AlphaFold3 code 'could potentially undermine Google's bargaining position in collaboration discussions with industry partners'.
It says that, even though the code will only be made available for academic use, 'questions remain how this restriction will be enforced'. Even if patent protection is granted, there may be difficulties in detecting infringement.
A strong position
The article argues that 'Google DeepMind's strategy of switching from an open source to a commercial model for AlphaFold' has put it in a strong position.
It makes the point that the earlier versions of AlphaFold 'ensured early and widespread adoption of its model', and 'provided Google with a mountain of free evidence for the utility of its product. It goes on to say that 'the initial strategy of publishing the code for AlphaFold and AlphaFold2 in full did not prevent Google from simultaneously filing patent applications for the technology'.
A reputational hit?
The article says that the failure to provide the code for AlphaFold3 has been a 'reputational hit' for Google Deep Mind, as might be any attempt to enforce IP. It is now a case of waiting to see 'how its leaders will balance the competing pressures of keeping academia onside, whilst securing an attractive IP position for commercial partners in the pharma industry'.
This is an exciting and evolving field, and we look forward to seeing how it continues to develop.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
