Scientists use machine learning to improve drug research

Scientists have succeeded in finding potential medicines faster with the help of machine learning. After training, the technology can better predict which molecules have desirable properties compared to conventional systems.

The machine learning technology was developed by scientists from the authoritative institute MIT. According to the researchers, a model was developed that was then trained with a dataset containing data from about 250,000 molecules, with different properties. The intent was for the system to use the data to predict how best to optimize molecules to serve as potential drugs.

MIT’s experiments show that after training, the machine learning model is better able to optimize molecules than existing systems; properties such as synthesisability and solubility were better achieved through machine learning. Even when the system was asked to find the best ‘base molecule’, it outperformed conventional systems.

In drug development, a basic molecule is usually chosen in the lab that contains desirable properties, such as binding to a specific receptor in the body. Subsequently, the basic molecule is further improved by means of a so-called lead optimization process, so that it acquires properties that are important for medicines, including the possibility to manufacture it easily, or to have it properly dissolved in liquids. After the necessary tests, the optimized molecule is then tested on laboratory animals before being tested on humans.