This AI system requires only a small amount data to predict molecular characteristics
The process of discovering new materials and drugs is usually a manual one, involving trial and error. This can take years and cost millions. Scientists use machine learning techniques to predict molecule properties, and to narrow down molecules that they will synthesize and test.
Researchers at MIT and MIT Watson AI Lab developed a new unified framework which can predict molecular property and create new molecules more efficiently than popular deep-learning methods.
Training is the process of showing millions of molecular structures to a machine learning model in order to teach it how to predict the biological or mechanical properties of a molecule. Large training datasets can be difficult to obtain due to the cost of finding molecules and the difficulty of labeling millions of structures by hand. This limits the effectiveness machine-learning methods.
Source:
https://phys.org/news/2023-07-ai-small-amount-molecular-properties.html
