Combining physics-based models and big data can lead to systematically better hypotheses.

Finding a reasonable hypothesis can be difficult when there are thousands of possibilities. This is why Dr. Joseph Sun II Kwon seeks to develop hypotheses that are generalizable and systematic.

Kwon, an associate professor in Texas A&M University’s Artie McFerrin School of Chemical Engineering, published research in Nature Chemical Engineering that combines traditional physics-based scientific models with experimental data to accurately predict hypotheses.

Kwon’s research goes beyond traditional chemical engineering. By combining the laws of physics with machine learning, his research has the potential to impact renewable energy, smart manufacturing, and healthcare, as outlined in his recent paper, “Adding Big Data to the Equation.” Outlined.

Deriving hypotheses from experimental observations typically requires a trial-and-error process. Kwon has developed a systematic framework that integrates expert knowledge and experimental data to create more efficient processes.

“The most impactful aspect of this research is that it bridges the gap between theoretical models and real-world complexity, creating a versatile framework for solving complex problems,” Kwon said. states. “This versatility means the potential benefits extend across a wide range of industries and can have a major impact on everyday life.”

Kwon said this research could lead to the discovery of new drugs by incorporating experimental data into these models. This hybrid modeling approach integrates biological knowledge and data to accelerate drug prediction.

“Developing new drugs is expensive and time-consuming,” Kwon said. “However, more sophisticated models can accelerate the discovery and manufacturing process. Leveraging simulation and machine learning reduces the need for expensive laboratory experiments, saves time, and The discovery of new treatments will be accelerated.”

His approach combines physically-based models with the flexibility of data-driven components that can adapt and modify predictions based on real-world experimental data.

Kwon plans to use these models as the backbone to simulate complex systems and capture fundamental physical phenomena that traditional physics-based models alone cannot capture.

“This methodology allows us to continuously estimate process parameters in parallel with the hyperparameters of data-driven components,” Kwon said. “Doing so allows the model to be applied to a wider range of situations, making it more versatile and capable of responding to new and diverse scenarios.”

“Purely data-driven models are insufficient to capture the complexity of these systems,” Kwon says. “By combining the two approaches, we can improve the efficiency and reliability of industrial processes essential to the production of everyday necessities such as energy, chemicals, and pharmaceuticals.”