A collaborative research team consisting of scientists from the MIT-Takeda Program has developed an innovative method that combines physics and machine learning to analyze the rough surfaces of particles in pharmaceutical pills and powders. The goal of this research is to improve the manufacturing process of pharmaceutical products by reducing subjectivity and enhancing efficiency.
When manufacturing pills and tablets, pharmaceutical companies need to separate the active pharmaceutical ingredient from a suspension and dry it. Currently, this process relies on human operators to monitor an industrial dryer, agitate the material, and visually assess when the compound has the desired qualities for compression into medicine. However, these subjective observations can lead to inconsistencies and errors.
To address this issue, the research team from MIT and Takeda collaborated on a project aimed at using physics and machine learning to categorize the rough surfaces of particles in pharmaceutical mixtures. They developed a technique called the physics-enhanced autocorrelation-based estimator (PEACE), which has the potential to revolutionize pharmaceutical manufacturing processes for pills and powders. By applying PEACE, efficiency and accuracy can be increased, leading to fewer failed batches of pharmaceutical products.
The reliability of pharmaceutical manufacturing and the reduction of time and compliance issues are crucial considerations in the industry. Failed batches or steps in the process can have serious consequences. Therefore, any improvements that enhance the manufacturing process and ensure product quality are highly significant.
The collaboration between MIT and Takeda is part of the MIT-Takeda Program, which focuses on leveraging the expertise of both institutions to address challenges at the intersection of medicine, artificial intelligence, and healthcare.
In traditional pharmaceutical manufacturing, determining whether a compound is adequately mixed and dried typically requires stopping an industrial-sized dryer and taking samples for testing. The research team at Takeda believed that artificial intelligence could improve this task and reduce production slowdowns caused by stoppages. Initially, the plan was to train a computer model using videos to replace human operators. However, the subjective nature of selecting videos for training proved to be a challenge. As a result, the team decided to utilize a laser to illuminate particles during filtration and drying, combined with physics and machine learning to measure particle size distribution.
By shining a laser beam onto the drying surface and observing the interaction between the laser and the mixture, the team was able to apply a physics-derived equation to characterize the particle sizes. Machine learning algorithms were employed to analyze the data. This approach eliminated the need to stop and start the process, making it more secure and efficient compared to standard operating procedures.
One of the advantages of this method is that it does not require a large amount of training data for the machine learning algorithm. The physics-based components enable the neural network to be trained quickly and efficiently, reducing the reliance on extensive datasets.
Currently, there are no inline processes in the pharmaceutical industry for measuring particles within a powder during mixing, except for slurry products where crystals float in a liquid. However, when a liquid is filtered and dried, its composition changes, necessitating new measurements. The PEACE mechanism not only improves the efficiency of the process but also enhances safety by minimizing the handling of potentially potent materials.
The implications of this research for pharmaceutical manufacturing are significant. It can lead to more efficient, sustainable, and cost-effective drug production by reducing the number of experiments required during the manufacturing process. Monitoring the characteristics of a drying mixture has long been a challenge in the industry, and the PEACE mechanism represents a substantial advancement in real-time particle size distribution monitoring.
The research team believes that this mechanism could have applications in other industrial pharmaceutical operations as well. Future developments may include training video imaging systems to replace laser measurements. Currently, the company is assessing the tool on different compounds in its lab.
The successful collaboration between Takeda and MIT's departments of Mechanical Engineering, Chemical Engineering, and Electrical Engineering and Computer Science has been facilitated by the MIT-Takeda Program. This collaboration has resulted in 19 projects over the past three years, all focused on applying machine learning and artificial intelligence to healthcare and medical challenges. The direct collaboration and real-time feedback from Takeda have allowed researchers to align their research with industrial processes, potentially shortening the timeline for translating academic research into practical applications.
