How end-to-end laboratory automation and AI are accelerating drug discovery

The next era of laboratory automation is no longer about sample movers or liquid handlers. With big data being the central focus, several new trends have emerged in this space, making laboratory automation better, more powerful, and more accessible. By Anita Ramanathan.

Automation goals once revolved around ‘getting more done’. Now, the goal is to build a ‘smart laboratory’ where a central platform connects and controls multiple instruments. Meanwhile, an information management system captures and stores all the data generated, making it easier for machine learning algorithms to identify patterns and inform future decisions. Worried this might be a little too much to pull off? Enter cloud-based R&D laboratories that have everything set up and ready to go. 

From AI-driven drug discovery to robot scientists, here are some of the latest trends in laboratory automation. 

Workflows with end-to-end automation  

Automation has visibly evolved from simply performing repetitive tasks to now intuitively fulfilling the entire workflow, from start to finish, better known as end-to-end automation. This is where different automation systems within a facility connect and communicate with each other to execute workflows autonomously. For example, in cell-based microscopy, the liquid handler, sample mover, incubator, and microscope are all centrally connected. When the sequence is triggered, the workflow initiates, advancing from one step to the next, all while capturing the data and sending it to a cloud-based server. 

A specially designed software is used to connect all these systems, plan workflow sequences, and monitor progress remotely, sometimes in real-time. Such an integrated ecosystem of automation can significantly eliminate the white space between individual workflows, resulting in a smoother, faster operation for the entire laboratory. 

“Our goal in drug discovery is to use automation as a tool to find drugs faster, reduce costs, and improve accuracy,” notes Dr Cathy Tralau-Stewart, Chief Scientific Officer at C4X Discovery, a company that combines drug discovery expertise and cutting-edge technologies. “We can now employ automation in almost every aspect of drug discovery – from finding and validating targets to running clinical trials. We need to leverage any tool – automation included – to make our processes more efficient, and hopefully, reduce attrition rates.”  

To experience immediate efficiency gains from end-to-end automation, without being weighed down by the technical minutiae of algorithms, it helps to collaborate with vendors specialising in automation software and hardware to customise and build out these sequences. The deployment of these advanced automation projects can range from a few months to almost a year based on the sophistication of the workflow. 

One such success story of end-to-end automation accelerating the drug discovery pipeline comes from the AstraZeneca iLab in Gothenburg, Sweden, where compound synthesis runs on automation. “What was once a manual, multi-step process in medicinal chemistry is now fully automated, from start to finish”, says Dr Michael Kossenjans, Head of iLab, AstraZeneca. “We set up, perform, and monitor the chemical reactions. Once complete, we can then isolate, purify, and quantify the crude product – all from a single platform.” 

The iLab team has successfully built a centrally connected automation system that can manage different instruments involved in the compound synthesis workflow, i.e, supplying the precursor chemicals, performing the analytical chemistry, and preparing the stock solution for screening. Using carefully crafted workflow sequences, small molecule compounds are now synthesised and purified autonomously. 

Centralising data and adopting good data practices 

Automation generates massive amounts of data. This can be good news or bad news – it all depends on the data practices enforced within the laboratory. Given the large volume of data being produced, laboratories need to have a digital infrastructure in place to manage it. 

Data gathered from automated workflows will need to be appropriately captured, annotated, and stored. Digital tools can be typically used to centralise and manage all the data from instruments, samples, inventory as well as end-to-end workflows. Instruments with their own vendor software platform connect to the central data repository through an application programming interface (API).  

Modern data management solutions specifically designed to support automation in laboratories are already equipped with all these necessary features. It is, however, important to standardise experimental processes and data capture practices across the company. This means, experiments are kept consistent, and for each experiment, the user records the same set of data points every single time. “A big part of my role as the data director is to ensure that datasets are gathered properly,” says Dr Yohann Potier, Director of Data Platform at Tessera Therapeutics, a company pioneering in a novel genome engineering technology. Our automation, computational and scientific teams work in close collaboration to uphold good data practices.”  

“There is immense value in data – that is where the intellectual property is,” adds Dr Potier. “In the era of automation, I would say a company’s data is now considered to be its biggest asset.” Cloud-based information management systems can be used to centralise data and make it accessible remotely in real-time, eliminating data silos between teams or departments. This stored data can then be retrieved in the future to power machine learning algorithms. But there’s one caveat – the available data needs to be ‘clean’. 

“As laboratories build their automation systems with the goal of using AI in the future, having a clear strategy for gathering and storing this data is an important step,” notes Anca Ciobanu, Theme Lead at Pistoia Alliance, a not-for-profit membership organisation aimed at bringing R&D leaders together for pre-competitive collaborations. “Changing data practices every so often or swapping platforms without a purpose decentralises the laboratory’s data, making it very challenging to make use of it later on, ”continues Ciobanu. “Implementing AI is only possible with centralised, good quality data.” 

Leveraging AI in small and big ways 

A 2021 report1revealed that AI will have the most impact on the pharmaceutical industry in the coming years, attracting almost 100 collaborations between pharma companies and AI vendors in recent times. Although AI is already being successfully used in tech industries and for image and speech recognition, it is now beginning to make strides in drug discovery as well. 

In laboratories with automated workflows, the data generated from a series of experiments can be used to train AI systems to learn the ‘rules’ of the experiment and infer new knowledge accordingly. AI is essentially the ability of a computer to mimic human cognitive functions, such as learning and troubleshooting. Machine learning, on the other hand, is a subset of AI. It uses data to learn and improve without any intervention, imparting intelligence to the computer.  

In a scientific setting, AI can be used to run simple tasks or to perform complex multi-step workflows. At the AstraZeneca iLab, AI is used in small and big ways. “We use an AI-based platform that provides us with suggestions on how to synthesise compounds”, says Dr Kossenjans. “In the past, to synthesise a molecule, a chemist would have to scribble down all the different steps to be performed. Now, we have algorithms where we simply input the desired molecule, and it shows us the steps required to synthesise it, along with protocols and literature references.” 

Dr Kossenjans continues: “Running a fully autonomous workflow with no human input wouldn’t be possible without AI because there needs to be a programme that can make decisions in our absence. Our automated workflows have an ‘if this happens, then do that’ feature. For example, if the pressure in the reactor vial gets too high, then stop and cool immediately. That’s AI at work.” 

The team is now building a more advanced AI model to make decisions on the experimental process itself, beyond the instrument. For example, if particular outcome wasn’t achieved, the AI will analyse what went wrong and automatically choose how to modulate the temperatures or solvent concentrations, and then repeat the experiment. The fundamental goal of AI in drug discovery is to progress experiments in a self-determining, independent manner to reliably achieve the desired output. 

The arrival of robot scientists  

What if your lab partner happened to be a robot? That’s what Dr Ross King, Professor of Machine Intelligence at Chalmers University of Technology, studies and develops. 

Dr King’s team built the world’s first robot scientist, named Adam, that successfully made a novel scientific discovery in yeast genomics without any human intervention. A robot scientist is essentially a computation system that is capable of generating its own hypothesis, performing experiments, analysing data, and repeating the entire cycle. Its brain is a collection of computers; its body, a set of laboratory robotic arms connected to incubators, liquid handlers, and other relevant instruments.  

And it’s the kind of lab partner that doesn’t stop until it finds an answer. 

“We equip the robot scientist with background knowledge. With that, it has an automated way of formulating hypotheses using machine learning”, explains Dr King. “It can then select different experiments to test each hypothesis. We optimised it for time and speed – so it would design experiments that are as cheap as possible and get to the answer as fast as possible. The robotic arms then carry out the experiment, providing results that the robot scientist can analyse. This cycle then continues until there is only one final theory that is consistent with the background knowledge provided.” 

After Adam came Eve, the second robot scientist from Dr King’s laboratory that can automate early-stage drug design. Eve can autonomously perform library screening, confirm hits, and generate leads. Currently, the next-generation robot scientist being developed in his laboratory is called Genesis, equipped with AI capabilities to design, plan, and execute thousands of parallel experiments in complex biological models. 

Simultaneously, Dr King is also working on the ultimate challenge – building an AI-driven robot scientist that can autonomously design and make Nobel Prize-quality scientific discoveries by 2050. 

While the idea of robot scientists taking over experiments in drug discovery may make some uncomfortable, when seen through the lens of shrinking the drug discovery timeline, there’s no tool better than machines, according to Dr King: “Machines are much faster than humans at deciding what compound to test next. They can analyse copious amounts of data from previous experiments and comb through millions of scientific publications. And they can do so more logically, more reliably, and better than humans ever can. I believe robot scientists have the potential to significantly improve productivity and efficiencies in the drug discovery process, and ultimately save time and costs.” 

Can AI and robots ever replace scientists in drug discovery? 

Over the years, laboratory automation has progressed in its capabilities and sophistication. Now, with AI-enabled troubleshooting and robotics, experiments can be autonomously performed with minimal human intervention.  

Understandably, some scientists worry if these machines might be smart enough to replace humans. While it’s certain that the role of a scientist within an R&D laboratory is bound to change in the near future, machines cannot replace scientific thinking, experts concur. 


“This concern stems from our reluctance to embrace change. Organisations that are investing in automation need to engage in open dialogue with their scientific team. It needs to be clearly communicated that automation will make it possible for them to work on more valuable and creative tasks that ultimately accelerate the drug discovery process. 

Also, many scientists trained in academic settings may not have had the opportunity to work alongside automation systems. Therefore, offering training opportunities to build their knowledge around AI can alleviate some of these concerns. Over time, with increased awareness, scientists will become more receptive to their changing roles.” – Anca Ciobanu, Theme Lead, Pistoia Alliance.

“Humans and machines are better together than they are alone. AI systems can already perform the complete loop of scientific research, including hypothesis development and data analyses. My contribution to science is the world’s first robot scientist that autonomously made a novel scientific discovery. Since then, my team has built more advanced robot scientists. 

AI systems have superhuman capabilities to remember, learn, and extract useful information from enormous datasets. I believe we need to explore this power of AI, especially if it can get us better, faster results in drug discovery.  

On the other hand, humans have a much better understanding of the overall goal of a project. Robots don’t get the ‘big picture’ as they can only solve a specific problem based on the knowledge we provide.” – Dr Ross King, Professor of Machine Intelligence, Chalmers University of Technology.

“My position on this is very much inspired by chess. A computer can be programmed to become a great chess player. However, if we teamed this computer up with a grandmaster, we now have an unbeatable combination. Scientific work can be significantly enhanced by combining the core strengths of AI and the human mind. 

We became scientists because we were curious about the world around us… about how things work. What if we applied that natural curiosity to not only explore what we’re studying, but also how we’re studying it? Having that openness and being willing to learn about AI and how to leverage it for better and faster research will be key.” – Dr Martin-Immanuel Bittner, CEO and Co-Founder, Arctoris.

“I don’t think robots can replace us, but they can definitely transform what we do and how we do it. This is an industry that cannot – and should not – sit still. Being in drug discovery requires embracing innovation.  We need to be willing to learn new and different ways of doing things.

By leveraging the speed of robotics, we can actually do better science – produce higher quality data, screen fewer compounds, and bring drugs to market faster. As for the fear of being taken over by robots… If scientists ultimately control and program the robots, then there’s actually nothing to worry about, right? Scientists can now focus on the science rather than routine tasks and can gain valuable insights from AI and machine learning.” – Dr Cathy Tralau-Stewart, Chief Scientific Officer, C4X Discovery.

“Machines can give us more freedom to creatively think about a scientific problem. Let’s take an example from the field of imaging. Histopathologists spend years in training learning about how cells look under a microscope. On the job, they used to spend hours every day examining a single image and drawing contours around cells. Now, a machine can do this task in seconds. But histopathologists haven’t lost their jobs. They now bring other strengths, such as adding new targets to the imaging protocols and reviewing data. 

In much the same way, when AI and machine learning take over some of the tasks, scientists can spend more time designing new experiments or scaling up existing workflows. Dr Yohann Potier, Director, Data Platform, Tessera Therapeutics.

Remote R&D labs and strategic partnerships 

For an individual laboratory, automating a single task can be rather straightforward but building an in-house robotic system that can accelerate the R&D process isn’t trivial. This is where outsourcing to or partnering with R&D laboratories with niche expertise and ready-to-go automation systems can come in handy.  

Cloud-based R&D laboratories are now emerging both in service-based start-ups, such as Emerald Cloud Lab and Strateos that provide remote-controlled R&D services to research groups, as well as inside global pharmaceutical companies to streamline internal projects across different teams.  Using a secure portal, a task triggered from any site in the world can now be completed at a dedicated R&D laboratory located remotely. “When we build our fully automated, cloud-powered R&D laboratories at AstraZeneca in a year or two, we want our scientists from around the world – in US, UK, China, Sweden and beyond – to be able to use them”, shares Dr Kossenjans. “No matter where the laboratory is physically located, experiments can be started from a user’s desktop.” 

Establishing strategic partnerships with drug discovery companies that have in-house expertise in data science and automation can also help accelerate the R&D pipeline. “We automate every step of the drug discovery process – target validation, hit ID, hit-to-lead, and lead optimisation”, says Dr Martin-Immanuel Bittner, Co-Founder and CEO of Arctoris, a tech-enabled drug discovery company that leverages robotics, data science, and machine learning to progress a pipeline of programmes in oncology and neurology.  

This new category of drug discovery companies, such as Arctoris, is entirely powered by data. “There are literally hundreds of AI companies out there, but only a handful that combines robotics, data science, and AI,” says Dr Bittner, who is a clinician-scientist by background. “We developed a fully automated laboratory architecture that allows us to generate structured, reproducible, high-quality data that is machine-learning ready by design. That means, we can power our AI’s decision-making process with this superior quality data, enriched with metadata and contextual information. We strongly believe that the key to successful drug discovery lies in having better data to make better decisions – and ultimately benefit patients worldwide.” 

Here’s how strategic partnerships work with data-powered R&D companies: Once you’ve established a partnership agreement, and both scientific teams have agreed on the milestones they would like to achieve together, the experiments are built using the vast library of automated protocols that the company, such as Arctoris, has already developed and optimised. When the project begins, experiments run in a robotic wet laboratory, generating data 24×7.  

Leveraging a fully automated R&D workflow has the potential to produce more than 100x the number of data points per experiment compared to industry standards, according to Dr Bittner. Plus, the data is automatically captured, annotated, quality-controlled, visualised and analysed. This allows the iterative ‘design-make-test-analyse’ cycle in R&D to be accelerated several-fold. 

Collaborating to serve a unified cause 

As the pharma industry navigates the unpaved path towards AI-driven drug discovery, sharing knowledge and collaborating with peers will be necessary. At the moment, each company, working independently, tackles AI-related problems on its own, unaware of the fact that dozens of others might have a similar challenge. But what if we were to start exchanging ideas and share our best practices with industry peers? Working together – and setting standards along the way – might just help us all get to our goals faster. After all, every company in the industry strives to achieve the same cause – bringing better drugs to the market, faster.  

“We learn as much from our missteps as from our successes,” says Ciobanu, who moderates in-depth discussions on topics such as automation, AI, and machine learning for professionals across the industry. “Presenting use cases at conferences and outlining AI strategies that worked, along with those that didn’t, can help companies learn from each other. That’s why at Pistoia Alliance we encourage open discussions between small and large pharma companies to facilitate knowledge sharing, brainstorm ideas, and provide access to information otherwise contained within companies.”  

“We’re also seeing better integration between automation vendors and software providers,” notes Dr Potier. “They’re now working together to provide APIs and open databases so lab-generated data can be easily converted to reports. This type of integration was nearly impossible a few years ago.” 

On another note, a critical issue resulting from automation that requires our immediate attention is the plastic waste generated from single-use consumables. “We perform thousands of experiments on our automated platforms, often in parallel, producing excessive amounts of plastic waste around the world, each day”, says Dr Kossenjans. “This is something that warrants our attention right now… while we build our automation systems. Because it doesn’t make sense to advance one domain of science while creating a new problem for another.” 

Drug Discovery World e-book – innovating in the lab 

About the author 

Anita Ramanathan is a science writer and award-winning speaker based in Bristol, UK. In her capacity as a science writer/editor at several digital publications, including NIH Research Matters, she has crafted dozens of stories buried under numbers and scientific findings. A storyteller at heart, Anita also delivers science communication workshops. 









Related Articles

Join FREE today and become a member
of Drug Discovery World

Membership includes:

  • Full access to the website including free and gated premium content in news, articles, business, regulatory, cancer research, intelligence and more.
  • Unlimited App access: current and archived digital issues of DDW magazine with search functionality, special in App only content and links to the latest industry news and information.
  • Weekly e-newsletter, a round-up of the most interesting and pertinent industry news and developments.
  • Whitepapers, eBooks and information from trusted third parties.
Join For Free