Referenced literature:
DeepMind Blog: A glimpse of the next generation of AlphaFold
Molecular Docking: A powerful approach for structure-based drug discovery
Tech Crunch: Isomorphic inks deals with Eli Lilly and Novartis for drug discovery
Researchers use AI-powered database to design potential cancer drug in 30 days
Dear readers,
Our topic today comes from the news this week that Isomorphic Labs signed partnerships with both Eli Lilly and Novartis, with $45M and $37.5M payments upfront, respectively. Both deals are geared towards discovery of small-molecule drugs. Isomorphic Labs, which aims to use artificial intelligence to aid in drug discovery, will receive this upfront cash and possible milestone payments down the road. If that sentence sounds familiar, you’re not alone. I, too, was reminded of a certain Israeli company (at which I’m currently employed) upon reading the story. So much so that I put together this little quiz to see if you can guess whether the following claims come from the website of A: Isomorphic Labs , or B: CytoReason . Answer key is at the end of the newsletter.
Who said it? Isomorphic Labs or CytoReason?
Statement 1: Bring data to life. From trial and error to predictable medicine.
Statement 2: We’re a digital biology company, here to redefine drug discovery with the power of artificial intelligence.
Statement 3: The traditional process of drug discovery is too long, too expensive and too risky. We believe AI can improve it – helping to get better drugs to the people who need them, and to treat and cure disease faster.
Statement 4: Our mission is to make actionable information accessible to everyone at the time it matters.
So what is this company and what is interesting about them? Isomorphic Labs was founded under the Alphabet umbrella in 2021 by Demis Hassabis, CEO of DeepMind. The idea was to leverage DeepMind’s AlphaFold technology to reshape drug development. To those who missed the story from a few years back, AlphaFold made major waves back in 2021 when it was released, showing capability to accurately predict 3D models of thousands of protein structures. As a side note, I remember the day that AlphaFold was released because it was a few days after my good friend Scott defended his structural biology PhD thesis, where over the span of five years he had managed to solve the structures of two or three complex proteins. I asked him what he thought about it and whether the tool had any future. He responded by saying he’s not sure but it seems like his PhD was now obsolete.
Since then, DeepMind and Isomorphic Labs have improved AlphaFold further. While the initial release of AlphaFold was key for identifying single chain proteins, current versions can predict much larger and more complex proteins. An example of how AlphaFold can help in improving the process of drug discovery is in how we determine interactions between ligands and proteins. Previously, this was done by taking several configurations or conformations of the ligand in the active site of a protein, and ranking them by a scoring function. AlphaFold bypasses the need for a reference protein, or suggesting how the ligand should bind, or even the need to test proteins that have been structurally solved. It allows us to predict the binding properties of proteins that we don’t even know the structures of. Pretty amazing.
AlphaFold takes in as an input a description of the biological assembly for both the protein and ligand, and optionally the sequence location of covalently bonded ligands, and outputs a prediction of the position of each atom. In one of their latest blog updates, they show that this approach works for proteins, nucleic acids, small molecules, ions, and modified residues. An example of AlphaFold helping in drug development came last January, where researchers were able to apply the tool to identify inhibitors for a pathway in hepatocellular carcinoma, just 30 days after target selection (!), by testing compatibility of the target with several structures.
While that is an example of what is potentially very exciting about AlphaFold, there is still room to improve. In a commentary in Science, limitations of the tool are addressed, noting that “[there is] still no substitute for actual experimental data”. AlphaFold isn’t perfect. The estimate is that between 7-20% of its structures contain residues in incorrect orientation. For better or worse, this error does not seem to be random. But even in this critique, the author notes the excitement surrounding the power of AlphaFold, and that at the very least it is an extremely effective hypothesis generating tool.
For Isomorphic Labs, this marks an exciting first step into the world of pharma and definitely a name to keep an eye on. As always, with brevity in mind, I left a lot of information out in this note. But feel free to take a look at the referenced literature at the top of the newsletter, or shoot me a slack if you’d like to chat more.
Wishing everyone a safe and enjoyable weekend! :)
Tommer
Answer key below!
Answer Key:
Statement 1: Bring data to life. From trial and error to predictable medicine.
Answer: CytoReason
Statement 2: We’re a digital biology company, here to redefine drug discovery with the power of artificial intelligence.
Answer: Isomorphic Labs
Statement 3: The traditional process of drug discovery is too long, too expensive and too risky. We believe AI can improve it – helping to get better drugs to the people who need them, and to treat and cure disease faster.
Answer: Isomorphic Labs
Statement 4: Our mission is to make actionable information accessible to everyone at the time it matters.
Answer: Trick question, this is from Theranos’ twitter bio 😮