Using our state-of-the-art AI models to power drug design
Pascal Savy joined Isomorphic Labs in December 2022 as a Research Leader in the Drug Design and Medical Research (DDMR) Team.
Before joining Isomorphic Labs I was already very interested in the potential applications of machine learning to drug discovery and had a keen interest in property prediction models as well as generative chemistry models. I was also very conscious of the limitations of methods available to the drug design process at the time, but felt the field was becoming increasingly mature.
When I started working at Iso, I was extremely impressed by the capability of the models developed since the company’s inception and how they allowed us to experiment with ideas and concepts. For me, the impact of the models was clear in a number of areas. For example, they could help us test new hypotheses for the binding of ligands to proteins of interest.
AI models to support a multi-disciplinary drug design team
The whole process of bringing new medicines to patients is very long and complex and success isn’t always guaranteed. Isomorphic Labs, however, is taking a unique approach in employing machine learning - we’re working on building models from first principles to answer the fundamental questions that can change the way we design drugs. Ultimately our goal is to make the process much more efficient, successful, and safer, leading to design drugs for challenging disease areas and that can result in fewer side effects for patients.
To achieve this, we’ve built a multi-disciplinary drug design team including medicinal chemists, computational chemists and biologists, working alongside machine learning researchers, machine learning engineers and software developers. Our culture places great emphasis on knowledge sharing between disciplines, fostering a culture of cross-pollination of ideas.
We also believe that in order to answer some of the most challenging questions about disease treatment and human health, we have to build a range of predictive and generative models that can support the work of our scientists across multiple expertise.
Although these models are extremely complex at their core, they are engineered in such a way that scientists without prior training in computer science or computational chemistry can use them to design ligands to interact with proteins. The usability of such tools is a very positive step forward, as different disciplines and ways of thinking will approach the same problem in many different and possibly unique ways.
Due to our constant knowledge sharing and the feedback loops we’ve built, the team is able to iterate on the technology to build models that are better able to impact the drug design process. I deeply respect the intellectual agility, openness and a curiosity to learn from other disciplines that this requires of my colleagues.
Working on challenges from different starting points
This process of developing and iterating Iso’s AI models for our team has played a critical role in our progress to date. Structural prediction of ligand-protein complexes is a very important component of the drug discovery process as it allows our scientists to understand intricate networks of interactions between proteins and small molecules. The nature of these interactions is in part responsible for the way such molecules inhibit or potentiate the inherent function of the proteins with which they interact.
In this regard AlphaFold 3, which was developed jointly by Isomorphic Labs and Google DeepMind, has been a complete game changer, especially for novel protein targets for which very little is known in the literature. AlphaFold allows a wide range of our team members to predict those interactions without any prior knowledge of these structures.
Many commentators in the industry have rightly remarked that it will take more than AlphaFold alone to radically change the way we discover new drugs and advance them to patients. AlphaFold is a hugely impactful innovation for biological research, but represents one piece of the drug discovery puzzle. Iso is addressing some of these surrounding challenges with its ecosystem of proprietary predictive and generative models that have been developed in-house by my colleagues in machine learning research. This pioneering AI technology is being rapidly iterated on, and is used daily here to support our research across the company.
Iso’s computational and medicinal chemists, for example, can formulate a hypothesis of binding between a small molecule and a protein, run a series of structural predictions using AlphaFold 3 and modify the chemical structure of such molecules virtually and also predict the way they can interact with that protein.
Additionally, through our suite of proprietary in-house models, our drug designers can predict their ability to efficiently bind to that protein as well as how it may behave in the body. They are able to explore and address their inherent solubility, permeability and metabolic properties which are essential for developing a successful drug. In this way, as the models get better, fewer molecules will need to be synthesised and tested in the lab, thus accelerating the whole process leading to identifying drug candidates.
Looking ahead
Beyond the small molecule drug design paradigm, these models have the potential to support new and innovative therapeutic approaches based on modulating protein-protein interactions as well as protein-nucleic acid interactions. Due to our access to huge compute power we can deploy these models at scale and interrogate multiple hypotheses simultaneously.
But the excitement of what lies ahead comes down to what I’ve learned about Isomorphic Labs since joining at the end of 2022 - The significance of Iso's technology is not only in its capabilities, the compute power and resources we have access to and the potential impact. It also lies within our team - our commitment to genuine collaboration, mixing the deep knowledge of our scientists with the expertise of our machine learning team, to develop the models and refine our methodologies and approach together. We believe this will lead to a reimagined drug design process, leveraging the benefits of pioneering technology at its core with the potential to impact patients worldwide.
.png)
.svg)
