Each step of drug discovery involves a lot of material and time costs. Computational approaches, especially artificial intelligence, are used to facilitate each of these steps.
So far, computational methods, especially machine learning, have made a significant contribution to the discovery of drugs.
One of the phases of drug discovery is "lead optimization". The purpose of this step is to increase the number of appropriate drug compounds to improve efficacy, reduce toxicity or increase drug absorption. Finding the right drug compounds requires an approach to extract specific properties from drug structures so that the right compounds can be selected from different structures.
Unfortunately, older algorithms for extracting properties from drug molecules could not be learned; But new methods based on machine learning have the ability to predict the biological activity of drugs.
In this seminar, one of the methods of extracting properties from pharmaceutical structures and the challenges facing this issue are examined. The main focus of this method is to extract features directly from molecular structures based on deep learning approaches. The results show that these methods have the power to learn structures in proportion to their biological activity.