September 12, 2018
Combining deductive quantum chemistry and inductive machine learning
RIKEN CSRS has successfully predicted NMR chemical shifts with high accuracy using exploratory machine-learning algorithms.
Prediction of matters such as the weather, crop and catch harvesting, and health have historically been a vital issue for humankind. The use of artificial intelligence such as machine learning is now being used on vast amounts of big data to develop algorithms that can inductively predict various events. Conversely, NMR data uses quantum chemistry theory for deductive prediction, but the error between the theoretical value and the measured value is large and requires a correction value.
A RIKEN CSRS research team has developed a method that combines deductive quantum chemistry theory with inductive machine learning methods, exploring 91 machine learning algorithms to learn from and correct these errors and predict chemical shifts with a high accuracy.
In the future, a combination of theoretical chemistry and machine learning are expected to be used with NMR data (such as the chemical shifts from this study) in fields such as materials science and informatics to predict material properties.
Chemical Science doi:10.1039/c8sc03628d
K. Ito, Y. Obuchi, E. Chikayama, Y. Date, J. Kikuchi,
"Exploratory machine-learned theoretical chemical shifts can closely predict metabolic mixture signals".
Jun Kikuchi; Team Leader
Kengo Ito; Postdoctoral Researcher
Environmental Metabolic Analysis Research Team