AlphaFold 3 has sent ripples through the Molecular Biology and Drug Discovery communities. AlphaFold 2 was already considered to have revolutionized the field with its striking accuracy in predicting individual protein structures.
However, it had limitations in predicting complex multi-protein interactions and their interactions with key molecules such as DNA, RNA, and small molecules. AlphaFold 3 addresses these gaps and offers enhanced capabilities to predict these interactions accurately.
For more insights into AlphaFold 3, read our in-depth article and learn about the tech’s advanced protein interaction modeling, limitations, impact on industries, and more.
Impact on drug discovery:
AlphaFold 3 increases the potential to identify new drug targets compared to the previous version by including interactions with DNA, RNA, and small molecules. All of these critical steps are part of most disease mechanisms. It also impacts Lead Optimization by predicting drug metabolism and potential toxicity more accurately.
These improvements can save millions of dollars and months in development. A more accurate identification of the drugs’ affinity, stability, and even predictions of bioavailability and toxicity means a reduced number of wet lab experiments and stopping a lot of undesirable candidates from progressing into early clinical development.
A gated leap forward:
AlphaFold was originally released under the quite permissive open source Apache License 2.0. This meant that anyone could modify and use it, including for commercial use. With the full source code made publicly available, plenty of people did just that, improving the model and adding features, some of which made it back to the model itself. AlphaFold 3’s source code and model will not be publicly released, which has surprised many in the research community.
DeepMind has provided a web-based access point where researchers can submit up to 10 structures per day. However, the outputs are limited, and it’s not possible to obtain protein structures bound to potential drugs.
Part of a larger strategy:
This shift aligns with a broader strategy to monetize the tool. Isomorphic Labs, DeepMind’s commercial arm for AlphaFold, announced earlier this year a strategic partnership with Eli Lilly and Novartis. These contracts include modest upfront payments of $40-45 million, and could potentially reach nearly $3 billion combined, based on milestone achievements and excluding any royalties. With the potential of the current iteration of the tool and the success of this commercial model, it is easy to understand the change in openness.
What’s next to come?
While AlphaFold’s achievements are impressive, it is not alone in the field. The limitations on research access could drive improvements in the dozens of tools now available, such as RoseTTAFold or HADDOCK. The key data used by AlphaFold, sourced from the Protein Data Bank, is publicly available, allowing other researchers to build on this foundation.
Given the detailed information provided in AlphaFold’s Nature paper, it won’t be long before similar tools emerge, potentially offering open-source alternatives.
If you have any questions, please contact us here. You can also email the author, Joao Guerreiro, at jguerreiro@prescouter.com
