Michael Siedler and team at AbbVie’s labs in Ludwigshafen, Germany use predictive analytics to figure out how molecules will behave. Furthering Knowledge with Every Experiment For children, good behavior can lead to extra screen time. For molecules, good behavior can lead to the discovery of a life-saving medicine. And whether you’re a parent or a scientist, the ability to predict this good behavior before it happens can save time, energy and effort. While parents haven’t been as lucky, scientists are closer to cracking the code: researchers at AbbVie are working on using artificial intelligence that can help predict how a particular biopharmaceutical active substance will behave. Using a field of mathematical simulations called predictive analytics in combination with a new intelligent laboratory automation solution, researchers will be able to further scientific knowledge with every experiment. Michael Siedler, research fellow and head of High Throughput Screening and Advanced Formulation Sciences at AbbVie’s Ludwigshafen, Germany-based labs, explains the implications of this technology for researchers and ultimately, patients. Why are predictive analytics so important to biologics research? Michael Siedler: In the field of biologics research, our scientists are constructing novel, non-natural molecules to hopefully enable better therapies. But non-natural molecules have one disadvantage as compared to natural ones: they have not undergone any "evolutionary" processes, which are important, among other reasons, for sufficient stability. Therefore, we need to conduct numerous experiments in the laboratory to retrospectively stabilize these new molecules. The problem is, modern biological therapies are very complex to develop, and in order to construct stable molecules, we need a lot of data – meaning this takes time and resources that can ultimately delay the development process. This is where predictive analytics could come in. In general, we can use data to predict certain things, such as Amazon determining which products could be of interest to me based on my consumer behavior. In our case, we want to predict the stability of certain manufactured molecules based on their structure, in order to carry out fewer and more targeted experiments in the laboratory. 'Big data' is not an exaggerated expression: In one single test module, 'Alfons' collects an average of 105,408 data points to determine the stability of biopharmaceutical active substances. So, how do we go about obtaining the necessary data? What does the actual process look like? Siedler: For about four years now, we have been working on a new development concept in our Ludwigshafen labs enabling Automated Liquid Formulation Screening, or “Alfons”. Alfons is a new kind of robot, or rather a whole assembly line of automated instruments built for the screening and analysis of biopharmaceutical molecules. The basic concept is to standardize, miniaturize and, ultimately, enable automation when screening these manufactured molecules. Put simply, this means a few things: 1) our tests are always performed under the same conditions, due to standardization; 2) much lower amounts of active substance are required than with conventional methods, thanks to the miniaturization of the tests, which makes us faster and more efficient; and 3) with every new test, we collect data that – after processing – is added to a constantly growing overall data set. Number three is the most important benefit that Alfons brings, because we gain more scientific knowledge during each run. In some cases, results can be transferred to other molecules. With this knowledge, we can ultimately bring therapies to patients even faster and more efficiently. Instead of the approx. 200 g of active substance required in the past, the system only needs approx. 20 g of active substance and generates approx. 1000x more data in an electronically processable format. How has Alfons changed your way of working? Siedler: Our way of working is very different today than it was even a few years ago – even simply in the way our team is constructed. Nowadays, we need specialists in the areas of IT, data and database handling, software integration and automation. This was not conceivable a few years ago, when our teams generally consisted of the laboratory manager and around three employees with pharmaceutical, chemical or biological backgrounds. In the past, the development of new active substances was often based on and adapted from experiences with other antibodies, which worked well in earlier experiments. To a certain extent, this was based on intuition and pattern recognition. But with predictive analytics, using the variety of data we generate, we should be able to predict how a new molecule will behave and under which conditions it is most stable. This is fundamentally changing our development processes. Ultimately, we are moving from a previously limited understanding of the molecule to a comprehensive one. This means we would be able to predict, for a completely new molecule, whether and under what conditions it is stable enough for human application – whether it is a "lobster” or "crayfish." What do lobsters and crayfish have to do with drug development? Siedler: Our protein designers are very creative. They gave certain antibody formats names like “lobster” and “crayfish” because of how their form resembles certain sea creatures. Using novel molecular combinations, the aim is to develop therapies that intervene more effectively and efficiently in disease processes in the human body. What’s next for Alfons? Siedler: Our next goal is to achieve the full automation of our tests. The better the data position for a particular molecular format, the better we can then supplement experiments with mathematical models and, ideally, replace them. To do this, we must pool our knowledge and process it in such a way that it is always available to our researchers. This way, we can develop more innovative therapies for the most serious diseases and make them accessible to patients as quickly and efficiently as possible.