Scientists have developed a ready-made deep learning model for predicting the pharmaceutical properties of cures- MolMapNet. The development is presented in an article for the journal Nature Machine Intelligence.
Researchers from universities in Singapore and China have developed MolMapNet, a new artificial intelligence (AI) tool. He predicts the pharmaceutical properties of cures by analyzing molecular concepts based on human knowledge.
While AI tools are generally good for recognizing spatially ordered images (such as images of objects), they don’t work as well with disordered data such as molecular properties. This impairs their effectiveness in the analysis of pharmaceuticals. Scientists have sought to overcome this limitation. The goal is to improve the performance of deep learning models for predicting pharmaceutical properties of cures.
The deep learning model was created in three stages:
- The first is a broad study of the internal relationships of molecular properties of more than 8 million molecules;
- The second is the use of a newly developed data transformation technique to display the molecular properties of pharmaceuticals in 2D images. Pixel layouts reflect the internal relationship between these properties. They contain important indicators of pharmaceutical properties that are captured using trained deep learning models;
- The third is teaching the MolMapNet tool to recognize 2D images and use them to predict pharmaceutical properties.
As a result, AI can capture specific pixel layout patterns that characterize specific pharmaceutical properties. It looks like that. how artificial intelligence distinguishes between men and women in the image, studying different gender characteristics.
Innovative AI does not require fine-tuning of parameters, which means it is available for non-specialists.