MIT and Japanese multinational pharmaceutical firm researchers have improved the pharmaceutical production methods for tablets and powders using machine learning and physics. Scientists devised a physics-enhanced autocorrelation-based estimator (PEACE) technique to classify particles with rough surfaces accurately. This method seeks to boost productivity and accuracy by decreasing unsuccessful batches of pharmaceutical goods by better identifying and analysing particle properties.
Allan Myerson, a Professor of Practice at the MIT Department of Chemical Engineering and one of the study’s authors, added, “Failed batches or failed steps in the pharmaceutical process are very serious. So, it’s a major deal if we can make the pharmaceutical production process more dependable, faster, and compliant.”
The research aimed to solve problems arising when healthcare, artificial intelligence, and medicine come together. A recent publication in Nature Communications discusses the approaches for making it far less subjective and more efficient.
In today’s practice, pharmaceutical businesses must separate the active pharmaceutical ingredient (API) from a suspension and then dry it to make the pills and tablets used to treat various diseases, conditions, and symptoms. A human operator must keep an eye on an industrial dryer, stir the material, and wait for it to reach the proper consistency before being compressed into medication. The operator’s keen eye is crucial to the success of the task.
It is customary to halt an industrial-sized drier and remove samples from the production line to test for proper mixing and drying in pharmaceutical production. Researchers hypothesised that AI could streamline the process and cut down on delays.
The study group’s first goal was to use video footage to teach a computer model to take the position of a human operator. However, deciding which films to utilise to train the model was still too subjective. So instead, the team used laser illumination during filtering and drying to determine particle size distribution by applying physical principles and computational analysis.
“We just shine a laser beam on top of this drying surface and observe,” Qihang Zhang, the study’s first author and a PhD student in MIT’s Department of Electrical Engineering and Computer Science explained.
The laser-mixture interaction is described by an equation derived from physics, and the particle sizes are characterised using machine learning. According to George Barbastathis, Professor of Mechanical Engineering at MIT and corresponding author of the paper, the task is safer and more efficient because it does not need pausing and beginning the operation.
Because physics permits fast neural network training, the machine learning algorithm only needs a few datasets to learn its job. To train the neural network effectively despite limited training data, “we use the physics,” as Zhang puts it. “You can get a good result with only a small amount of experimental data.”
Particle measuring inline techniques are exclusively employed in the pharmaceutical business for slurry products (where crystals float in a liquid). Its particles cannot be measured when a powder is mixed. Converting slurries into powders requires fresh measurements since the liquid’s composition changes as it is filtered and dried. The authors argue that the reduced need for handling potentially dangerous chemicals because of the PEACE mechanism makes the procedure safer overall.
The implications for the pharmaceutical industry might be huge, allowing for more efficient, sustainable, and cost-effective medication production by lowering the number of tests that manufacturers must do. Unfortunately, the time it takes for academic findings to be implemented in manufacturing is often prohibitive. However, there is optimism among experts that closer cooperation will reduce this time frame.