When medical companies manufacture pills and tablets that treat any number of ailments, aches, and pains, they need to isolate the active pharmaceutical ingredient from the suspension and dry it. The process requires a human operator to monitor the industrial desiccant, agitate the material, and watch for the compound to take on the correct qualities of pressure in medicine. The functionality is highly dependent on operator feedback.
Ways to make this process less subjective and more efficient is a topic of conversation Nature Communications The paper was authored by researchers at MIT and Takeda. The paper’s authors devise a way to use physics and machine learning to classify the rough surfaces that characterize particles in a mixture. This technology, using a Physically Augmented Autocorrelation-based Estimator (PEACE), can transform pharmaceutical manufacturing processes for pills and powders, increasing efficiency and accuracy and resulting in fewer failed batches of pharmaceutical products.
“Failed batches or failed steps in a pharmaceutical process are extremely dangerous,” says Alan Myerson, professor of practice in MIT’s department of chemical engineering and one of the study’s authors. “Anything that improves pharmaceutical manufacturing reliability, reduces lead times, and improves compliance is a big deal.”
The team’s work is part of an ongoing collaboration between Takeda and MIT. The MIT-Takeda Program was launched in 2020. The MIT-Takeda Program aims to leverage the expertise of both MIT and Takeda to solve problems at the intersection of medicine, artificial intelligence, and healthcare.
In the pharmaceutical industry, determining if a compound has been sufficiently mixed and dried usually requires stopping an industrial-sized dryer and taking samples from the manufacturing line for testing. Takeda researchers believe AI can improve the job and reduce downtime that slows down production. Originally, the research team planned to use the videos to train a computer model to replace a human operator. But deciding which videos to use to train the model is still very subjective. Instead, the MIT-Takeda team decided to illuminate the particles with lasers during filtration and drying, and measure the particle size distribution using physics and machine learning.
“We just shine a laser beam over this drying surface and observe it,” says Qihang Zhang, a doctoral student in MIT’s Department of Electrical Engineering and Computer Science and first author of the study.
The physics-derived equation describes the interaction between the laser and the mixture, while the machine learning characterizes the particle sizes. The process does not require stopping and starting the process, which means the entire job is safer and more efficient than standard operating procedures, according to George Barbastathis, professor of mechanical engineering at MIT and corresponding author of the study.
The machine learning algorithm also does not require many sets of data to learn its function, because the physics allows for rapid training of the neural network.
“We use physics to make up for the lack of training data, so we can train the neural network in an efficient way,” Zhang says. “Only a tiny amount of empirical data is enough to get a good result.”
Today, the only inline processes used to measure particles in the pharmaceutical industry are for slurry products, where crystals float in a liquid. There is no way to measure particles within a powder while mixing. Powders can be made from a slurry, but as the liquid is filtered and dried its composition changes, requiring new measurements. In addition to making the process faster and more efficient, using the peace mechanism makes the job safer because it requires less handling of potentially high-strength materials, the authors say.
The implications for drug manufacturing could be significant, allowing drug production to be more efficient, sustainable and cost-effective, by reducing the number of trials companies need to conduct when making products. According to Charles Papageorgiou, director of the Process Chemistry Development Group at Takeda and one of the study’s authors, monitoring drying mixture properties is a problem that has long plagued the industry.
“It’s a problem a lot of people are trying to solve, and there isn’t a good sensor out there,” says Papageorgiou. “That’s a pretty big change, I think, in terms of being able to observe, in real time, the particle size distribution.”
Papageorgiou said the mechanism could have applications in other industrial pharmaceutical processes. At some point, laser technology may be able to train video imaging, allowing manufacturers to use a camera for analysis instead of laser measurements. The company is now evaluating the tool on different compounds in its lab.
The results came directly from a collaboration between Takeda and three MIT departments: mechanical engineering, chemical engineering, electrical engineering, and computer science. Over the past three years, researchers at MIT and Takeda have worked together on 19 projects focused on applying machine learning and artificial intelligence to problems in healthcare and the medical industry as part of the MIT-Takeda Program.
Often, academic research can take years to translate to industrial processes. But the researchers hope that direct collaboration will shorten that timeline. Takeda is within walking distance of the MIT campus, which allowed the researchers to run tests in the company’s lab, and real-time feedback from Takeda helped the MIT researchers structure their research based on the company’s equipment and operations.
Combining the expertise and mission of both entities helps researchers ensure that their experimental findings will have real-world implications. The team has already filed two patents and has plans to apply for a third.
(tags to translation) MIT-Takeda software