The Science of Scale-Up

Tom Kalil

Typically, researchers who are using AI to accelerate the pace of scientific discovery are focusing on the earliest stages of the innovation process, such as “can I find a novel chemical that has some desirable set of properties?” Researchers have rarely focused on the “science of scale” e.g.,what is the best way to produce tons of a chemical?

Linden Schrecker, CEO of the UK-based startup Solve Chemistry, is trying not only to solve the problem of novelty, but also scale - “What’s the best way to make a ton of a novel and useful chemical?”  Recently, I interviewed Linden to learn more about the “science of scale-up”, which I believe will be critical to accelerating the pace of innovation.

What problem is Solve Chemistry tackling, and how did you get interested in this problem?

We are tackling how we make chemicals efficiently and quickly at scale. Our vision is a future where once someone has made a milligram of a new chemical, they could efficiently make a million milligrams - a ton - the very next day if they want to.

I became interested in this area after my undergraduate degree in Chemistry at Oxford. Although we were good at designing and making new molecules (amplified by more recent development of companies such as Isomorphic Labs), the field was not good at collecting data which allowed these chemistries to be easily repeated by other researchers or taken to an industrial scale - especially not in an efficient or sustainable manner. This motivated me to pursue an interdisciplinary PhD at Imperial College London sponsored by BASF at the intersection of chemistry, chemical engineering, and machine learning.

My PhD focused on how we can most efficiently collect data from chemical reactions – focussing on types of data industry struggled to produce but would provide real value. My interest in this problem only grew as I interacted with more chemical companies, and I could better understand the most important problems that needed to be solved. As interest in automation and AI in other sectors grew, the lack of data we had in the production of drugs and agrochemicals only became more apparent.

Why is this an important problem for the chemical industry?

Developing a new chemical – be it a drug or an agrochemical, a semiconductor or a plastic – comes with risk. At every stage of development, the aim is to reduce the risk of failure and having good data early on helps to reduce this risk. However, there are also trade-offs to be considered: people want to assess how different decisions would affect the cost, speed, robustness, and sustainability of their process. Having more data allows trade-offs to be accurately assessed and derisks decision making.

AI is finding more and more success in other parts of the drug development pipeline, increasing hit-to-lead success rate and designing more targeted treatments, relying on data collected by researchers over the last 100 years driven by regulation. We do not have this quantity of data for process conditions, which is limiting the usefulness of AI. The increased success rates in early parts of the development pipeline is putting more pressure on the production stage. To reap the rewards of these early successes, we need to collect data fast to solve this growing bottleneck.

Why should people outside of the industry care about progress on the “science of scale-up?”

We want more drugs (healthcare) and agrochemicals (food security) to be produced – and we want to protect the planet while doing so. It is great to be able to design the best molecules for the job, and to test them on a small scale, but these molecules will only have societal benefits if they can be produced efficiently at scale. Currently our development cycles are slowed down by not being able to produce materials when we need it, this wastes money and time that could be used to develop more new drugs and agrochemicals – and limits the impact that new science has on people’s lives.

What’s a goal related to the science of scale-up that industry and academia could pursue?

A lot of great research is coming out of academic chemistry labs, however this research is rarely performed with scale-up in mind. Better data collection can solve this, which will  smooth the path to industrial uptake. Although there are large datasets on distinct chemical reactions, specifics on the conditions these are performed under are rarely reported or explored – these are crucial.

It is in everyone’s interests for this data to exist, so why doesn’t it? The incentive structures of academia and pharma are such that the faster your next paper or next drug can be released the better.  As a result, collecting or publishing extra data for any research result is not prioritised. This type of data is also challenging to collect efficiently by traditional means and requires interdisciplinary teams to handle. SOLVE Chemistry is well placed to male this happen.

What would we need to do to make progress on this goal?  For example, you’ve noted that one of the reasons we have made comparatively little progress on using AI to optimize manufacturing processes for chemistry is that there is very little publicly available data.   What steps could we take to change that?

We can make a big difference on our own by producing data and building predictive AI models to allow more efficient and sustainable scale-up. However, this could be significantly amplified if we can make the collection and sharing of high-quality process data part of the culture of chemical R&D.

We want to take the best new chemistry from academic labs and produce the data that makes it possible to immediately use in an industrial setting. This will allow a significant shortening of timelines for new research being used safely and efficiently in the industrial production of drugs and agrochemicals. The outcomes of this will be more new chemicals, more sustainably produced, faster.

To achieve this, we believe that by targeting cutting-edge research from academic labs and partnering with them to quickly collect high quality data in our systems, we could supercharge their uptake in industrial production. The production of this data, its publication with high prestige scientists, and its industrial use will also inspire others to collect more data – thus building the base needed for the worthwhile application of AI in the “science of scale-up”.