Discovery via machine learning algorithms requires large datasets of training data. We had problems predicting the properties of molecules and generating new ones. This can be solved by machine learning and deep learning approaches. However, these approaches require a large amount of training data to solve this.
The goal of researchers is to accelerate the discovery and materials development of new drug molecules. To tackle this problem, researchers at MIT discovered a way to predict the molecular properties of molecules using small datasets. A team of researchers created a machine learning model that automatically learns the language of molecules. It is known as “molecular grammar”. This technique is more convenient as it applies to smaller datasets. All information and grammars from small datasets are used. Get molecules with similar structures and understand the similarities between these molecules. The system understands the laws governing molecular similarity through reinforcement learning. The accuracy and f1 score of the model are even closer to achieving that goal. Molecular grammars are broadly classified into two parts. The first part is called metagrammer and the second part is called hierarchical approach.
This new molecular grammar system outperformed some machine learning models. Better results are obtained with very small datasets compared to datasets used to predict molecular properties via machine learning models. This is a powerful technique and can be applied to graph-based datasets as well. This makes it viable for both regression and classification approaches. So, to push the research further, the research team cut the training dataset in half and found better results. This was one of their remarkable achievements.
This method has been used in various fields, such as the prediction of physical properties such as the glass transition temperature. The research team hopes to apply the molecular grammar model to 3D molecules and polymers. Models based on molecular grammars lead to discovery of new molecules and prediction of their properties.
Please check paper and MIT article.don’t forget to join 26,000+ ML SubReddit, Discord channeland email newsletterShare the latest AI research news, cool AI projects, and more. If you have any questions regarding the article above or missed something, feel free to email us. Asif@marktechpost.com
🚀 Check out 800+ AI tools in the AI ​​Tools Club
