When medical companies produce pills and pills to treat various ailments, aches and pains, the active pharmaceutical ingredient needs to be separated from the suspension and dried. The process requires a human operator to monitor industrial dryers, agitating the ingredients and ensuring the compounds are of the right quality for compression into drugs. The work is highly dependent on the operator’s observations.
How to make that process less subjective and much more efficient is a recent theme. Nature Communications Papers written by researchers at MIT and Takeda. The authors of this paper use physics and machine learning to devise a method to classify the rough surfaces that characterize particles in mixtures. Using a physics-enhanced autocorrelation-based estimator (PEACE), this technology can change tablet and powder drug manufacturing processes, improving efficiency and accuracy, and reducing failed batches of drugs. .
Alan Meyerson, professor of practice in the MIT Department of Chemical Engineering and one of the study’s authors, said: “Anything that improves pharmaceutical manufacturing reliability, reduces time and improves compliance is important.”
The team’s work is part of an ongoing collaboration between Takeda and MIT. The MIT-Takeda program aims to leverage the experience of both MIT and Takeda to solve problems at the intersection of medicine, artificial intelligence and healthcare.
In pharmaceutical manufacturing, it is usually necessary to shut down industrial-sized dryers and take samples from the production line for testing to determine if compounds are properly mixed and dried. Researchers at Takeda Pharmaceutical thought artificial intelligence could improve tasks and reduce stoppages that slow production. Originally, the researchers planned to use video to train computer models to replace human operators. However, deciding which video to use to train the model still proved to be too subjective. Instead, the MIT-Takeda team decided to laser the particles during filtration and drying and use physics and machine learning to measure the particle size distribution.
“You just put a laser beam on top of this dry surface and watch it,” says Qihang Zhang, a Ph.D. student in the Department of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology and the study’s first author. .
Machine learning characterizes particle size, while equations derived from physics describe the interaction of lasers and mixtures. According to George Barbastathis, professor of mechanical engineering at MIT and corresponding author of the study, the process does not require stopping or starting the process. This means the whole process is safer and more efficient than standard operating procedures.
Also, machine learning algorithms don’t need a lot of datasets to learn their job. This is because physics allows rapid training of neural networks.
“We use physics to fill in the gaps in training data so that we can train neural networks efficiently,” says Zhang. “A very small amount of experimental data is enough to get good results.”
Currently, the only in-line process used for particle measurement in the pharmaceutical industry is a slurry product in which the crystals are suspended in the liquid. There is no way to measure the particles within the powder during mixing. Powders can be made from slurries, but filtering and drying the liquid changes the composition, requiring new measurements. In addition to making the process quicker and more efficient, using the PEACE mechanism also makes work safer, because less potentially very potent substances need to be handled, the authors say. .
Reducing the number of experiments a company needs to perform when manufacturing a product can have a significant impact on pharmaceutical manufacturing, making it more efficient, sustainable and cost-effective. Monitoring the properties of dry mixtures is a problem the industry has long grappled with, said Charles Papageorgiou, director of Takeda Pharmaceutical’s process chemistry development group and one of the study’s authors. says.
“This is a problem many people are trying to solve, but there are no good sensors,” says Papageorgiou. “I think this is a pretty big change in terms of being able to monitor the particle size distribution in real time.”
Papageorgiou said the mechanism could also be applied to other industrial pharmaceutical operations. At some point, laser technology may be able to train video imaging, allowing manufacturers to use cameras for analysis rather than laser measurement. The company is currently working on evaluating the tool against various compounds in the lab.
This result is a direct result of Takeda’s collaboration with three departments at MIT: Mechanical Engineering, Chemical Engineering, Electrical Engineering and Computer Science. Over the past three years, MIT and Takeda researchers have collaborated on 19 projects focused on applying machine learning and artificial intelligence to healthcare and medical industry problems as part of the MIT-Takeda program. .
It can often take years before academic research is translated into an industrial process. But researchers hope that direct collaboration will shorten that timeline. Because Takeda is within walking distance of his MIT campus, researchers can set up tests in the company’s labs, and real-time feedback from Takeda helps MIT researchers improve the company’s facilities and operations. helped build research on.
Combining the expertise and mission of both entities allows researchers to see that their experimental results have real-world implications. The team has already filed two patents for him and plans to apply for a third.