SandboxAQ is a company that many pharma and biotech professionals will be keeping an extremely close eye on in the coming years.
Having SandboxAQ emerged as an independent company following spin-off from Alphabet in 2022, the startup is quickly carving its path as a leader in next-gen pharma and biotech solutions through AI and quantum technology innovation.
But what exactly is SandboxAQ offering to the pharma and biotech space? To answer this, we spoke with Andrea Bortolato, Vice President of Drug Discovery, who shared valuable insights into the company’s groundbreaking approach.
“AI machine learning is completely changing the way we think about drug discovery,” he explains. “There is really a huge expectation of what we can achieve now with these new methods, but in reality, there are some challenges that are unique in our field.”
Bortolato shines a light on limitations in the applicability domain, explaining that SandboxAQ’s core vision is to overcome this by using advanced simulation and physics-based methodologies.
According to him, the combination of cutting-edge AI and a quantum-level understanding of systems is crucial in developing the large quantitative models (LQMs) that pharma and biotech companies can depend on for drug discovery and development.
Some of the key insights from the discussion include:
• Revolutionary LQMs: How current machine learning and physics-based solutions can be used to create transformative LQM strategies.
• Accelerated discovery: The potential to reach clinical candidates with the highest probability of success, in the fastest possible way.
• Data confidentiality: How to prioritize data confidentiality when building an optimized, client-specific LQM solution.
Tune into full discussion for a truly unique insight into the potential use of LQMs, and their ability to transform drug development and drug discovery in the pharma and biotech industries globally.
Rebecca Williamson: Hi, there. My name's Rebecca Williamson, I'm the publisher of Fierce Biotech. And I'm here today with Andrea Bortolato, VP of Drug Discovery at SandboxAQ. Andrea, thank you for joining me.
Andrea Bortolato: Thank you. Thank you for the opportunity.
Rebecca Williamson: So, before we begin, can you tell me a little bit about yourself and about SandboxAQ?
Andrea Bortolato: Yeah. So, I'm leading the team working closely with clients in pharma, biotech, and academic groups. Really try to ensure the technology that Sandbox has been developing in the last few years is really impactful and optimal for the type of problem in drug discovery or drug development that they're facing. And before joining Sandbox, I worked in pharma, Bristol Myers Squib, Schrödinger, and biotech as well, like hepatitis therapeutics in the UK.
Rebecca Williamson: So, jumping in, can you describe how AI LQMs can be deployed in lead optimization and structure-activity relationship modeling?
Andrea Bortolato: Right. So, AI and machine learning is kind of changing completely the way we think about drug discovery, the way we do drug discovery. And there is really a huge expectation of what we can achieve now with this new methods. But in reality, there are some challenges that are unique of our field.
For example, the applicability domain of this machine learning AI solution could be limited by the data available. So, how can we be confident when we are trying to solve a new problem, like a first-in-class type of drug discovery project, that we have enough understanding of the problem and the target to really be able to deploy this AI solution in an optimal fashion?
So, the vision of Sandbox is kind of try to expand the applicability domain of this method beyond the training set available. Using simulation and physics-based method. So, this optimal combination of AI and kind of understanding at the quantum level of the system, is providing a large quantitative model strategy that is beyond what a simple AI solution can deliver.
So, this is a game changing kind of a new viewpoint on the problem where we try really to understand from a quantitative viewpoint what could be the most important type of experiment to improve the AI solution in [inaudible 00:02:23] to lead, but also in finding and potentially reaching the clinical candidate with the highest probability of success, but also in the fastest possible fashion.
The reality is these large quantitative models are changing really the way we are thinking about drug discovery, and they really provide a unique, optimal kind of a framework for the medicinal chemists to make the right decision. It's allowing them to take all the information available, ensuring that the decision is resulting in the best AI framework that could provide the best possible compound.
It's really asking the question, can AI learn physics? Can AI learn chemistry and biology? If we are able to reach that point, then we can really go beyond the applicability domain and the limit of the training set, because there will be key principle that AI will understand when you're making that prediction.
Rebecca Williamson: So, how can you transform current standard ML and physics-based solutions into an LQM strategy? And how can it be integrated with the existing process and traditional approaches?
Andrea Bortolato: So, the overall vision is really try to provide almost super-power to the chemistry team and the drug discovery team. There are a large number of available data publicly and potentially from the client corporate database. We want to ensure that all this information, everything is available in their hand. The key is try to, for every single data point, understand the value and potentially also the level of error or potential risk that that particular data point can bring when you make a prediction.
So, it is really combining an approach that explore and exploit the data in an optimal fashion following a Bayesian type of strategy. So, when you decide what experiment to do next, the idea is it could be useful to improve the AI solution and maybe it can be potentially game changing, but there is higher risk, if you will. On the other hand, it's that it could be that you are very confident on the quality of the prediction, but probably that prediction will be much closer to what you already know.
So, the idea is to bring all the expertise in the team together with all the data available and create a unique framework that is really including end-to-end the possible optimal strategy. Then the chemist is still the key decision maker in the process. And all his expertise, and his experience, and the team still play a crucial role, but the reality is we want to expand and provide a full landscape of opportunity to the chemist that can use effectively to make the right decision.
So, it's really kind of bringing those key machine learning AI solution to the next level and providing a full framework that address every single possible aspect of the problem at the same time. Because drug discovery is a multi-parameter optimization process. So, it's difficult even for a normal, even expert in the field to be able to look at a molecule and take in account all the different endpoints that need to be optimized in parallel.
It's not just affinity, right? There is permeability, [inaudible 00:06:12], PK. So, what could be valuable often for one endpoint could be decremental for another aspect. So, you improve affinity, but maybe the molecule is more lipophilic and less soluble. So, how do you find that balance and think in parallel on all these different kind of directions? AI is extremely powerful at that. They can take everything and take in account all the different endpoints and try to optimize in a synergistic way the workflow. And Gen AI, in particular, is really helping and game-changing in this space. So, you can have an AI solution, create a molecule in the pocket of the protein, and in parallel be able to co-fold the complex and try to optimize using reverse diffusion, affinity, and [inaudible 00:07:03] properties.
Rebecca Williamson: Now, when you're working with a partner to build an optimal LQM solution, how do you ensure data confidentiality and how client data are used?
Andrea Bortolato: Right, so this is a great, great point. So what we're doing is really trying to be a partner for the client, expanding their research team. You can imagine our model to be similar to what Palantir is doing for other kind of companies. So we are trying to work with them. First of all understanding the key problem and they are the expert in this area, in this disease space. They exactly know the problem and the challenges that we need to solve.
So we work closely with them to build this workflow that is optimal for the type of question we want to answer. And we have a toolbox with different parts that we will use. So together with them, we start to build this workflow that is really the best possible strategy to follow. And we use the data from the client in two different way. One is to ensure we have the best model from an AI viewpoint, but also too, using retrospective validation to have cycle of optimization of this workflow.
And their input is always extremely valuable because they probably already try a lot of different things and they clearly know what is the key challenge that we need to address and what are the gaps.
So my team in Rediscovery is working with the research team, SandboxAQ, bringing into the game expert in quantum chemistry, machine learning, AI, computational chemistry, medicinal chemistry, but also kind of expert in software engineering and cloud deployment. So we really create a solution that fills those gaps. We prove it retrospectively, and then we deploy it in a prospective fashion to really prove success. And success is defined by the client based on experimental data that are proving what is the requirement to access the chemical space with a clinical candidate.
So when another threat discovery company, we do not have a pipeline of molecule, so there is no competition with the client. We only provide the best possible AI solution that could be deployed in their infrastructure or in the cloud to solve their problem. And we provide target exclusivity to avoid any type of problem. And we SandboxAQ is also leading the quantum security kind of work with banks. So we are really in an optimal position to guarantee the level of security pharma and biotech really needs.
Rebecca Williamson: So is there anything you'd like to share with us before we close out?
Andrea Bortolato: I think this is an extremely exciting kind of time for drug discovery and drug development. There is really a need for game changing solution. And it's amazing to see how every single pharma and biotech really want to change the way they're doing drug discovery. There is a unique opportunity here to really do things in a different fashion and the vision of Sandbox is focus on the technology, create the best possible technology, and then enable every single pharma and biotech to reach the clinic and at the end help patient where there are unmet medical needs.
Rebecca Williamson: And that's a perfect place to close out. Thank you so much for joining me today.
Andrea Bortolato: Thank you. Thank you for the opportunity.