Tag

Molecular Properties

All articles tagged with #molecular properties

science-and-technology2 years ago•4 min saved

"Self-Taught Machine Learning Models Identify Molecular Properties"

Biomedical engineers at Duke University have developed a new method called yoked learning to improve the effectiveness of machine learning models in identifying molecular properties for potential therapeutics or materials. By pairing a teaching machine learning model with a student model, the technique, known as YoDeL, outperformed or matched the accuracy of active deep learning systems while being much faster. This approach could enhance the efficacy of deep neural networks and help discover new drugs and drug delivery solutions.

via Phys.org|

#active-learning #biomedical-engineering #deep-learning

science-and-technology2 years ago•5 min saved

"Revolutionary AI System Predicts Molecular Properties with Minimal Data"

Researchers from MIT and the MIT-Watson AI Lab have developed a unified framework that uses machine learning to predict molecular properties and generate new molecules using only a small amount of data for training. Their system has an underlying understanding of the rules that dictate how building blocks combine to produce valid molecules, allowing it to generate new molecules and predict their properties in a data-efficient manner. The method outperformed other machine-learning approaches on both small and large datasets, accurately predicting molecular properties and generating viable molecules with fewer than 100 samples.

via Phys.org|

#ai #data-efficiency #drug-discovery

science-and-technology2 years ago•4 min saved

Advancing AI in Molecular Research: Predicting Properties with Minimal Data

Researchers from MIT and the MIT-Watson AI Lab have developed a new machine learning framework that can predict molecular properties and generate new molecules more efficiently than existing approaches. The system uses a molecular grammar to understand the rules of how building blocks combine to produce valid molecules, allowing it to predict properties and generate viable molecules with a small amount of data. The method outperformed other machine learning approaches and accurately predicted molecular properties even with datasets of fewer than 100 samples. The researchers aim to use this approach to speed up the discovery of new molecules and extend its applications beyond chemistry and material science.

via MIT News|

#drug-discovery #machine-learning #mit