ADME properties of molecules are important for the features of compounds as a drug. If molecules will have poor solubility and chemical instability it will lead to low bioavailability. It means molecules will be less effective. Also, patients will need to take them many times per day.

 

Drugs that absorb poorly when taken orally must be administered in some less desirable way, like intravenously or by inhalation. Another important feature of new compounds is toxicity.

​

Scientists need to make sure molecules that they suggest as a drug candidate that will be safe and not toxic. 

Machine learning methods are useful tools in prediction of ADME and toxicity properties. In this project we implement and investigate the performance of the machine learning model using the Random Forest algorithm.

 

This model was able to predict solubility and toxicity properties of chemical compounds. We implemented this model using Python language and modern Python libraries such as TensorFlow and PyTorch. We trained the model with experimentally identified solubility of 1128 compounds   and toxicity data of 1491 drug compounds. After training we validated and tested the performance of the machine learning model on the test set of experimental data. Machine learning that we developed demonstrated sufficient accuracy.