By Chris Lo
Big pharma has increasingly been making use of robotics to automate specific processes, such as drug screening, anti-counterfeiting and manufacturing tasks. This market is on the rise, with robots becoming faster and more effective while taking up less space in busy laboratories.
But what if robots didn’t have to fit into the laboratory? What if a robot could be the laboratory? The next wave of pharmaceutical robotics seems to be moving towards joining up all the smaller tasks handled by robots into an integrated, automated method of drug delivery, with advanced artificial intelligences coming up with their own hypotheses, before testing compounds on assays, building on lessons learned and delivering novel therapeutics. In short, a totally robotic drug development process.
It’s a concept which some see as controversial or even faintly threatening, especially in an area as sensitive as pharmaceutical development. Professor Ross King, a robotics and computer science expert at Manchester University, doesn’t see it that way.
“In the future I think the whole process can be automated, from saying this is the disease we want to work on, make me the assays, go ahead and screen compounds, make new compounds…the whole thing can be automated,” he claimed.
King has dedicated his work over the last few years, initially at Aberystwyth University and now at Manchester, to developing robots with the intelligence to automate scientific research.
Adam and Eve: robotic researchers
The culmination of the work of King’s team has been the creation of two robots, Adam and Eve. Adam, the first of the team’s robotic systems, was designed to conduct experiments in the field of yeast genetics. Eve, the newer and more advanced of the two robot researchers, was created to automate the more immediately applicable process of drug screening and design.
“[Eve] integrates what are normally considered three separate parts of the drug screening pipeline,” said King. “It integrates the screening part, the hit confirmation part and the QSAR [quantitative structure-activity relationship] learning part. These are normally three separate processes in early stage drug design, and we integrated them into one robotic system. The intelligence part is the last part, doing the QSAR learning.”
Eve isn’t the humanoid robot which one might imagine. In fact, the system is closer in design to a self-contained laboratory, with laboratory equipment brought together by robotic arms allowing it to perform multiple research tasks without having to be reconfigured. Using a base of applicable knowledge and a form of logic that allows it to create hypotheses and test the validity of those hypotheses, Eve can test thousands of compounds against an assay to achieve a specific result.
The QSAR process is key to Eve’s intelligence. “This is what you do to learn the relationship between the chemistry in the molecules you’re interested in and how well they work in your assays,” King explained.
“So it’s like a mathematical formula that translates the structure into activity. People use machine-learning to learn these; what we’ve done is integrated that with the testing of new compounds, which is not normally done.”
According to King, the natural next step for Eve is to imbue the system with the ability to synthesise its own compounds for screening purposes, a common practice among human researchers. But this depends on securing much-needed funding.
King is hopeful that big pharma will be impressed enough by the steps made so far to invest in the next one. “I think a problem in the past has been that you need a large capital investment in robotics, so why would you use that to test what some would consider exotic ideas? But I think now we’ve got enough evidence to show that it’s a sensible approach.”
Advantages of automated research: systematic efficiency
Even at this early stage of experimentation with automated research, it’s clear that systems like Eve could make a serious impact on the efficiency of pharmaceutical development. For example, robots could streamline the compound screening process by starting to analyse hits during, rather than after, screening.
“When Eve is screening it starts thinking about what it’s seen in its hits,” King explained.
“So instead of blindly testing more and more compounds, it chooses the compounds from its library to test its hypotheses about what’s important in the structure of the molecule to make it active. And these compounds are chosen to produce the maximum amount of information about this function, and they’re also trying to find better compounds.”
Given the waste of time and money which can be incurred as a result of tiny oversights in drug development, the systematic approach of robotic system also lends itself to greater efficiency. Robots are naturally suited to performing functions such as ultra-precise pipette tasks, and according to King they make fewer mistakes “by orders of magnitude” than humans.
King also noted that the impeccable records kept and the explicit expression of all completed tasks could minimise or eliminate this kind of error, as well as easing the data transition when pharma companies merge and have to find a way to combine their records.
The most important pharmaceutical contribution of robotic research systems, in King’s opinion, will be their ability to conduct the thousands of experiments that will be necessary to improve the base-level biological knowledge that makes drug development possible.
“Human cells have got tens of thousands of genes, proteins and small molecules all interacting in a complicated spatial and temporal way,” he said. “It’s horribly complicated, and that’s just one cell.
“Unless we understand these things, a lot of these systemic diseases that the pharmaceutical industry is trying to treat will never be dealt with. Only with more robotics and automation will we ever be able to do enough experiments to understand the biology. Cancer is very complicated, [as is] obesity…psychiatric diseases are even more complicated. That’s what the pharmaceutical industry has had problems with.”
Knowledge, logic, intelligence: the characteristics of robotics
But what are the mechanical underpinnings which make these achievements possible? We humans have an intuitive understanding of the nature of knowledge from a human perspective, which is based on a mix of memory, creativity, natural curiosity and a host of other factors. Robotic ‘brains’ are wired in a completely different way, with an entirely different set of strengths and weaknesses.
“Adam and Eve have knowledge about the biology, and they have to reason,” explained King.
“They’re not generally intelligent – they don’t understand they’re doing drug design, they don’t know the structure of a DNA helix – all these basic things that human beings know. So they’re very specialised, but in that specialised area they know a lot – more than any human could know, because there are a lot of facts out there that you could reason with, which also can be used as knowledge.”
Simple deduction, the basis of mathematics and computer science, isn’t enough for a robot researcher because the concept of deductive reasoning (the sum of true premises must also be true) doesn’t allow for the logical leap required for discovery. But while deduction alone isn’t suitable for novel ideas, the logical principles that the likes of Adam and Eve rely on (abduction and induction) hold common ground with a fictional character famed for his deductive skills.
“It’s a bit like the reasoning that Sherlock Holmes uses,” King claimed. “What fact could possibly explain these observations I’ve made? Who could be the murderer?”
King said that the types of reasoning to which computers and robotic systems are suited could make them unparalleled scientists. “Science is a bit like a game of chess, where computers can do very well.
“It’s not like an open-ended social problem, which human beings have evolved to deal with. So I suspect computers can already make better choices about which compounds to test than experts. No one’s ever done a head-to-head test, but I suspect the computers would win. Science is not really our speciality.”
Can AI be trusted?
It’s unsurprising that the notion of robots conducting vital medical research in the future is worrying to some observers. After all, King himself admitted that robots are inflexible in their application of knowledge and have no conception of the context of their task. Doesn’t this make robots inherently untrustworthy?
However, without contextual knowledge there is also no motive, no intent – two things that humans have in great supply. “One of the advantages of science as a problem for AI over general intelligence is that in science, if you do an experiment, nature doesn’t lie,” said King.
“If you do experiments to find out whether something turns red if you add acid, it either does or it doesn’t. If you repeat it enough, you’re going to find out. But if you ask someone something, you can never be sure why they’re saying what they’re saying. The machine doesn’t have any motive, and neither does nature, so it seems.”
In many ways, artificial intelligence and the concept of machine-learning is a natural extension of some of the modern pharma industry’s more advanced data-driven drug design models. King describes adaptive trial design, whereby clinical trials can be tweaked part-way through without losing statistical integrity, as “a related idea”, for example. Automation could simply be seen as a means of assigning the menial bulk of scientific discovery to specialist systems, allowing humans to concentrate on their own strengths, like strategic, intelligent choices.
According to a recent review by the British Medical Journal, “a culture of haphazard publication and incomplete data disclosure” is having a significant negative impact on clinical development and knowledge. With this in mind, it’s not unreasonable to wonder if humanity is the ideal workforce for the everyday, nuts-and-bolts process of drug development.
Furthermore, despite our fears, the advance of machine-learning and artificial intelligence is well-positioned to start picking up the slack in laboratories worldwide over the next few decades. Who knows, robot researchers might even end up being more trustworthy than their human counterparts.