Revolutionary AI-Laser Combo from MIT Transforms Medicine Production

Researchers from the MIT-Takeda Program have revolutionized the production process of pharmaceutical tablets and powders by integrating physics and machine learning. The method, known as PEACE, leverages lasers and machine learning to evaluate particle size distribution, thereby enhancing efficiency, minimizing production failures, and rendering the process more sustainable and economical.

A unified team of researchers from the MIT-Takeda Program has harnessed physics and machine learning to investigate rough particle surfaces in pharmaceutical tablets and powders.

An assembly of engineers and scientists from MIT and Takeda are employing physics and machine learning to advance the manufacturing methodologies of pharmaceutical tablets and powders. The objective is to optimize efficiency and precision, leading to a reduction in unsuccessful product batches.

During the manufacturing process of pills and tablets to treat various illnesses and discomforts, it is necessary to separate the active pharmaceutical ingredient from a suspension and dry it. This procedure involves a human operator overseeing an industrial dryer, stirring the material, and monitoring the compound to acquire the suitable properties for compression into medicine. This task heavily relies on the operator’s observations.

The focus of a recent article in Nature Communications, written by researchers from MIT and Takeda, is on methods to make this process less reliant on subjective judgment and considerably more efficient. The authors propose a method of using physics and machine learning to classify the rough surfaces that are characteristic of particles in a mixture. The technique, which implements a physics-enhanced autocorrelation-based estimator (PEACE), could revolutionize the manufacturing procedures of tablets and powders, boosting efficiency and accuracy and reducing unsuccessful batches of pharmaceutical products.

Allan Myerson, a professor of practice in the MIT Department of Chemical Engineering and one of the study’s authors, says, “Failed steps or batches in the pharmaceutical process are quite grave. Anything that augments the reliability of pharmaceutical manufacturing, saves time, and boosts compliance is hugely significant.”

Researchers’ innovative endeavor is part of a sustained collaboration between Takeda and MIT, initiated in 2020. The MIT-Takeda Program strives to utilize the expertise of both MIT and Takeda to solve challenges at the crossroads of medicine, artificial intelligence, and healthcare.

In the pharmaceutical production realm, determining if a compound is sufficiently mixed and dried usually requires halting an industrial-scale dryer and extracting samples from the manufacturing line for testing. Takeda researchers speculated that artificial intelligence could optimize this task and minimize disruptions that impede production. Initially, the research team intended to utilize videos to train a computer model to supplant a human operator. However, deciding which videos to use for training the model remained too subjective. Consequently, the MIT-Takeda team chose to light up particles with a laser during filtration and drying, and evaluate particle size distribution utilizing physics and machine learning.

Qihang Zhang, a doctoral student in MIT’s Department of Electrical Engineering and Computer Science and the study’s primary author, says, “We merely project a laser beam on top of this drying surface and observe.”

An equation derived from physics describes the interaction between the laser and the mixture, while machine learning determines the particle sizes. The procedure doesn’t necessitate halting and resuming the process, meaning the entire task is more secure and efficient than the usual operating process, according to George Barbastathis, a professor of mechanical engineering at MIT and the corresponding author of the study.

The machine learning algorithm requires minimal datasets for learning its task, as physics enables speedy training of the neural network.

Zhang explains, “We leverage physics to compensate for the scarcity of training data, so that we can efficiently train the neural network. A tiny amount of experimental data suffices to obtain a satisfactory result.”

Currently, the only inline processes employed for particle measurements in the pharmaceutical industry are for slurry products, where crystals float in a liquid. There’s no existing method for gauging particles within a powder during mixing. Powders can be created from slurries, but when a liquid is filtered and dried its composition alters, necessitating new measurements. The authors assert that besides making the process quicker and more efficient, the PEACE mechanism enhances safety as it necessitates less handling of potentially highly potent materials.

The implications for pharmaceutical production could be momentous, allowing drug production to be more efficient, sustainable, and cost-effective by reducing the number of experiments companies must conduct when creating products. Monitoring the characteristics of a drying mixture is a problem the industry has long grappled with, according to Charles Papageorgiou, the director of Takeda’s Process Chemistry Development group and one of the study’s authors.

Papageorgiou says, “Many people are trying to solve this issue, and there isn’t an effective sensor out there. This is a significant step forward in being able to monitor particle size distribution in real-time.”

Papageorgiou suggests that the mechanism could be employed in other industrial pharmaceutical operations. In the future, laser technology might train video imaging, allowing manufacturers to use a camera for analysis instead of laser measurements. The company is currently assessing the tool on different compounds in its laboratory.

The findings are a result of a collaboration between Takeda and three MIT departments: Mechanical Engineering, Chemical Engineering, and Electrical Engineering and Computer Science. Over the past three years, researchers at MIT and Takeda have collaborated on 19 projects centered on applying machine learning and artificial intelligence to issues in the healthcare and medical industry as part of the MIT-Takeda Program.

It can often take years for academic research to translate into industrial processes. However, researchers are optimistic that direct collaboration could expedite this timeline. Takeda is conveniently located within walking distance from MIT’s campus, enabling researchers to conduct tests in the company’s lab, and real-time feedback from Takeda assisted MIT researchers in shaping their research based on the company’s equipment and operations.

The integration of expertise and mission from both organizations aids researchers in ensuring their experimental findings will have practical implications. The team has already applied for two patents and plans to file for a third.

Reference: “Extracting particle size distribution from laser speckle with a physics-enhanced autocorrelation-based estimator (PEACE)” by Qihang Zhang, Janaka C. Gamekkanda, Ajinkya Pandit, Wenlong Tang, Charles Papageorgiou, Chris Mitchell, Yihui Yang, Michael Schwaerzler, Tolutola Oyetunde, Richard D. Braatz, Allan S. Myerson and George Barbastathis, 1 March 2023, Nature Communications.
DOI: 10.1038/s41467-023-36816-2.

Frequently Asked Questions (FAQs) about PEACE method

More about PEACE method

• MIT-Takeda Program
• Nature Communications article: Extracting particle size distribution from laser speckle with a physics-enhanced autocorrelation-based estimator (PEACE)