We took the time to speak with Stephan van Banning, chief technological officer (CTO) of solid state research instrumentation developer, Technobis Crystallization Systems, to discuss how AI applications for crystal imaging analysis fit in among this global trend.
BPR: What are the present challenges in crystallization image analysis?
For decades, crystallization researchers hunting the most stable morphology for different active pharmaceutical ingredients (APIs) faced an arduous challenge. They would conduct one experiment at a time, consuming a large working volume of starting material and struggling to observe the reaction with a probe before manually interpreting visual data.
The need for better visualization has encouraged technological development, namely camera installation into crystallization research instruments. By collecting pictures at regular intervals, researchers can observe the crystals in their dissolving and growth processes and ultimately say something about the size, shape, and distribution of crystals throughout. However, obtaining useful, good-quality pictures is not a simple task.
The environment inside a crystallization chamber is dynamic. Crystals are constantly growing or shrinking, dissolving and reappearing, and every image taken is different. As such, lighting must be expertly manipulated to ensure that every object in frame is in focus and detectable against the background.
Lighting is not the only issue with conventional camera imaging. In a 2D image, some shapes will inevitably stand in front of others. Distinguishing crystal outlines from shapeless masses is no easy feat.
BPR: How does this relate to AI?
This is where AI software can help. Extensive database training, consisting of annotating and designating countless images, enables the detection engine to count, measure and categorise any crystal within the reaction chamber.
But why is this important? What advantage is conferred by knowledge of the real-time microscopic developments of crystal morphology?
To researchers aiming to learn about the exact conditions for formation of the most stable crystalline forms, the data yielded by this image analysis represents the way forward.
BPR: What has been the impact of AI in crystallization research?
Finding a stable, bioavailable form of a molecule, one which can be scaled up to industrial production levels, is a pivotal junction in the drug discovery pathway. Nucleation, crystal growth, particle size distribution (PSD) and agglomeration are all crucial parameters to investigate.
Every year, millions of dollars are wasted when drug molecule candidates are, after months of investigation, ruled unviable due to physical or biological properties.
Advanced screening tools using AI image analysis empower scientists to make better decisions, earlier in the process. By gathering information about the optimal temperature and stirring conditions, researchers know exactly how to achieve the crystal form that will constitute the basis for the next stage of experiments.
BPR: Could you tell us about Technobis Crystallization Systems and its software?
An essential feature that Technobis Crystallization Systems uses in this next stage is feedback control. Its Crystalline V2 takes one image per second in each of the eight reaction chambers and, once the desired conditions are achieved, the software is able to control the conditions in each chamber to direct the crystallization process.
In some experimental instances where screening instruments have been connected to a robotic platform, an 24/7 automatic system that loads/unloads samples and determines the next stage of screening without human input has illustrated one possible future of the pharmaceutical research process.
As there are currently no commercial AI software packages available for this purpose, companies like Technobis Crystallization Systems are investing in producing proprietary image analysis engines. To help research groups to stay at the cutting edge of the field, the software often has a tool through which users can further train the database to recognise unexpected or new results.
Equipped with strong foundations on which to make decisions about the optimal research avenues, pharmaceutical scientists can navigate the murky waters of drug development in the smoothest and most efficient way.
Crystallization studies, powered by artificial intelligence, can yield high quality data simply. From there researchers can find constructive answers about the quickest, most cost-effective ways to bring medicines to those who need them.