Finding new medicines is complex. The traditional way of searching for drug molecules involves a lengthy and intensive period of optimisation cycles making and testing molecules, as well as manually reviewing vast amounts of literature and data – which can be a time consuming and expensive process. Artificial intelligence (AI) can transform this process and has the potential to revolutionise the way drugs are designed and synthesised. By harnessing AI, AstraZeneca is actively driving the change from traditional medicinal chemical approaches to a fully autonomous process by re-examining molecular drug design and synthetic chemistry methods. We talked to scientific experts at AstraZeneca: Ola Engkvist, Section Lead for computational chemistry in Gothenburg, Sweden, Clive Green, Head of Global Sample Management and Garry Pairaudeau, Head of Hit Discovery about their work in AI, their published article in Future Medicinal Chemistry and the implications of their findings on the use of AI in drug discovery.
A faster route to new drug-like medicines
Clive Green: I’m a medicinal chemist by training and have worked at AstraZeneca for 18 years so I know only too well that traditional computational approaches for evaluating molecular data can be augmented by allowing us to learn, unconstrained, from previous human designs. AI can help chemists explore vast chemical space, which can be as high as approximately 1060 molecules, on a timescale that can influence the design of molecules for our projects.
We’re turning to AI as a “complex data” processing tool and machine learning platform. By using AI, we have the ability to learn quickly from datasets and patterns. Crucially, we can then build knowledge and confidence to influence scientific decision-making on a drug discovery level. AI ultimately makes tackling complex data possible.
Ola Engkvist: Computational chemistry is my field and I lead a 15-strong team which has embraced machine learning to create and synthesise molecules. It’s hardly surprising that the use of AI technology in drug discovery is gathering interest and momentum when the molecular design phase can be reduced from two to three weeks to the same number of days. The time spent on discovering and refining a potential drug can in theory be reduced by two thirds, from three years to just one year. Significant advances in machine learning and deep learning techniques, which may have seemed like something from science fiction decades ago, are quickly becoming today’s reality.
Deep learning architecture and machine learning is complex. But, put simply, machine learning is like teaching a computer to learn a piece of music using algorithms. By providing the notes, the computer can learn the ‘piece’ and generate new music independently through deep learning. Translate this concept to science and it’s really exciting. We’re now in a position where we can train computational AI tools to use existing molecular structures and create new molecular structures.
Garry Pairaudeau: Scientists have used computers for years to help guide the synthetic routes needed to make drug-like molecules, mainly to access literature information. Historical efforts to build retrosynthesis tools have not been widely accepted, and lack the sophistication of a trained synthetic chemist. Now with AI, we are starting to see the use of deep learning to create rules on the fly and in some ways mimic the ways that chemists think about the synthesis process. I am very optimistic that we will soon see the development of valuable synthetic planning tools, which will save time for the chemists, enabling them to plan syntheses better.
Breaking new ground in publications
Clive Green: We’re proud to say that AstraZeneca is one of the industry’s leading experts in this area. In our article, we bring together years of our research to show how unifying AI and chemistry can improve drug discovery. In the article,1 the first stage addresses the design question - “What are we going to make?” Computational AI tools generate molecules and score them against parameters that drive the research projects. Using an iterative approach, the computer can then learn which compounds to make. The next stage, which involves synthetic chemistry planning, tackles the question of “How are we going to make it?” Using the compounds identified, machine learning and neural network algorithms can generate the molecular structures that hold the most potential activity.
Enhancing our expertise with collaborations
Clive Green: Our reputation in our use of AI in drug discovery encourages external partners to collaborate with us. As a team, we are building on our industry-leading position in this field, but we can develop further and faster by teaming up with external experts.
Ola Engkvist: We’re advancing algorithmic processes, through collaborations with the University of Bern and University of Bonn and exploring large scale applications of deep learning as part of the H2020 project ExCape together with University of Linz and Alto University in Helsinki. In the past two years, scientists at AstraZeneca have published over 20 scientific publications, showcasing improvements in algorithmic processes. Members of the team also actively take part in algorithmic competitions to drive benchmarks and push the boundaries when generating new algorithms.
Garry Pairaudeau: The academic outreach we have is impressive. We have a fantastic environment where students can work with us – sometimes for a week, sometimes for a year. This approach, driven by Ola and his team, has given us such a strong scientific base and ensures knowledge is cycled between industry and academia. We also make the most of informal collaborative opportunities both inside and outside the company. My colleagues Michael Kossenjans and Paul Harper have driven the development of our automated chemistry laboratory at AstraZeneca. It all started with a weekend hackathon at Munich University, where around 50 people worked on our stand to bring automation to chemistry. Truly inspired, we staged our own hackathon for a week with 16 colleagues from around the world. One of our people from Boston brought a crate of robotic arms, dumped it in the middle of the lab and off we went, building equipment and writing software. Actually, it makes me feel quite emotional talking about it. This is different from the way big companies usually work with detailed project plans. There is obviously a place for this, but on this occasion, we were spontaneous: humanistic input into a robotic future.
The dawning of an AI future
Use of AI in medicinal chemistry fits well within our 5R Framework approach.2 AI can be incorporated within all stages of the framework, from finding the right target through to the right safety and right patient. With the 5Rs embedded into the way we work and already a four-fold increase in research and productivity over five years, AI delivers the potential to drive this success even further in the future.
Clive Green: We are aiming to use AI in drug discovery to advance the chemistry lab of tomorrow – what we see today, but in the future fully automated with robotics. However, we will still need to synthesise and test compounds manually. This gap in development is what differentiates medicinal chemistry AI from other sectors, such as gaming or chess. As industry-leading experts in this field, we need to push and invest in advancing AI and automated chemistry to reduce the iterative cycle time when generating, validating and testing a high-quality compound.
This is ambitious and there are always concerns about how this will impact on jobs. Will humans still be needed? The answer is ‘yes!’
Garry Pairaudeau: There will always be roles for smart, innovative scientists in drug discovery. We are dealing with very different challenges from 25 years ago, no longer limited to small molecules; proteolysis-targeting chimeras (PROTACs), RNA, complex peptides and oligonucleotides have changed the drug discovery landscape for a medicinal chemist. We need people with experience who can apply their knowledge to the future with an everchanging scientific landscape.
AI does have cost savings benefits, but it also democratises knowledge, enabling scientists to make better creative decisions and solve problems. We currently have to rely on the most experienced people. Now, by combining our collective experience using AI, we can present 资讯rmation 更多 clearly across a disparate knowledge base, bringing far greater experience into the decision-making process.
Clive Green: I believe the role of the chemist will evolve and focus more on strategic scientific decision-making and solving the really challenging problems with creativity and imagination, rather than traditional testing and synthesis routine work. Chemists will always be needed to perform highly skilled experiments that are difficult to automate, but AI will allow us to focus on cutting edge experiments and adding high quality data to research projects. The possibilities of using AI are exciting for chemists – a transformation in drug design and synthesis is truly underway and I can’t wait to see what the future holds.
1. Clive P Green, Ola Engkvist, Garry Pairaudeau. The convergence of artificial intelligence and chemistry for improved drug discovery www.future-science.com/doi/full/10.4155/fmc-2018-0161
2. Morgan P, Brown DG, Lennard S, Anderton M, Barett JC, Eriksson U, Fidock M, Hamren B, Johnson A, March RE, Matcham J, Mettetal J, Nichols DJ, Platz S, Rees S, Snowden MA, Pangalos MN. Impact of a five-dimensional framework on R&D productivity at AstraZeneca. Nat Rev Drug Discov. 2018;17(3):167‒181. doi: 10.1038/nrd.2017.244.
Ola Engkvist Associate Director, Discovery Sciences Computational Chemistry, IMED Biotech Unit LOL下注网站
Ola Engkvist, a computational chemist, is the Section Lead for computational chemistry in Gothenburg, Sweden, specifically focusing on discovery functionality within the IMED Biotech unit. He leads a 团队 of approximately 15 people, which includes scientists, Masters and PhD students, and post-docs. The 团队 focuses on utilising machine learning to create molecules (molecular synthesis) and synthesise them (synthetic chemistry). The 团队 also focuses on chemical automated Hit and open innovation activities. Dr Engkvist gained his PhD at the University of Lund, Sweden and completed his post-doc in Cambridge, and began working at AstraZeneca in 2004.
Clive Green Executive Director and Head of Global Sample Management, IMED Biotech Unit LOL下注网站
Clive Green is the Head of Global Sample Management and is a medicinal chemist by training. Clive has worked at AstraZeneca for the 18 years, across a variety of areas, including oncology and cardiovascular. He currently manages a leading group of scientists who are responsible for the curation and production of small molecule and human biological samples in the IMED Biotech Unit. He is also responsible for developing and implementing research strategies, including AI use.
Garry Pairaudeau Head of Hit Discovery, IMED Biotech Unit LOL下注网站
Garry Pairaudeau is currently Head of Hit Discovery in the IMED Biotech Unit, responsible for generating high quality chemical starting points for projects working with colleagues and collaborators worldwide. He has worked for 25 years at AstraZeneca, and is a chemist by training with a PhD and post-doc in synthetic chemistry. He chairs the Global Chemistry Leadership team, and is responsible for chemistry strategy. He has been very active in building automation and machine learning capabilities, with the aim of speeding up the evaluation of starting points.