Keep up-to-date with the latest IP insights and updates as well as upcoming webinars and seminars via HLK’s LinkedIn page, or simply subscribe to our updates.
Welcome back to the AI Observer, HLK’s newsletter tracking the AI legal landscape.
In the latest issue, our digital-health team talk through what is going on with bioinformatics patent filings and the application of AI to solve the knotty problem of protein structures. We also provide an update on the ongoing trial in Getty Images v Stability AI.
Did you miss the previous issue of the AI Observer? Check out all our previous issues and other AI updates here.
Whipping biology’s math into shape (possibly gaussian)
Biological data is renowned for being complex. Take Diseases. Diseases are often not the result of one mutation in one gene but rather a pattern of mutations across various genes. Detecting cancer in slide or radiology images depends on a myriad of features in these images. Often in these fields, researchers and diagnosticians are looking to find patterns in biological data which are too complex for our human brains. Enter bioinformatics, the field of research in which biological data meets maths and computation. This field is currently enjoying a boost as AI tools – highly suited to identifying incredibly complex patterns – become more readily available.
Unsurprisingly, we are in the midst of a surge in patent applications in bioinformatics at patent offices around the world. This includes “classical” bioinformatics, where the maths is derived by humans, and AI bioinformatics where the maths is inferred or developed by AI.
Read the full article here.
Legislation watch
There’s really only one show in town at the moment for UK practitioners: Getty Images and Stability AI.
With the trial ongoing at time of writing, there are yet no conclusions to be drawn. However, it has been interesting to see trademarks move clearly to the forefront of this case. Indeed, and disappointingly for fans of clear case law, it appears that Getty have dropped its main copyright claims, likely in view of Stability AI’s argument that any alleged actions took place outside the UK – Getty have cited lack of clear evidence on this point. This means a key point will go undecided, although Getty’s claim for secondary copyright infringement remains at issue. Interesting too are Stability AI arguments that that the images in the trial do not represent real life, that the watermarked images were effectively coerced into being by Getty to prove a point. Photos of Jürgen Klopp holding a football shirt have lead us all to think about how many times Jürgen Klopp has been pictured holding a football shirt. All in all, it’s riveting stuff and even without the main copyright claims, the judgement should make interesting reading.
AI application spotlight | Diffusion: not just for hairdryers
It used to be that to solve a protein structure for drug discovery, scientists would sit in a basement shooting x-rays at crystals made using cat hairs (we are not fibbing: one of our digital health team was one practitioner of this dark art). To the relief of cats everywhere, we now have the CASP competition which pits teams of structural biologists and AI whizzes against each other to try and solve protein structures using AI. AlphaFold 3 is the most recent iteration of these from Isomorphic labs. Launched last year, it can compute not just single protein structures but also the structure of other biological molecules, e.g. RNA.
So what did they do to make this happen?
The answer is it uses a diffusion model (See here for more detail).
In particular, the software extracts clues for an unsolved protein structure by looking at similar amino acid sequences (specifically the evolutionary changes in them which points to co-evolving amino acids which are likely in contact) and already solved protein structures (that could also help point to interacting parts of the protein if the unsolved protein is similar enough to them). This provides a go-to guide to the likely amino acids in contact in the protein structure. The diffusion bit then starts with amino acids randomly placed in space, and using the guide, denoises the randomness using the guide and its training.
The result is a new AI model that can solve the structure of various types of biologically important molecules.
The cats are pleased!
This is for general information only and does not constitute legal advice. Should you require advice on this or any other topic then please contact hlk@hlk-ip.com or your usual HLK advisor.