Virtual screening of prodrugs

Nanjing University Of Posts And Telecommunications | China

Wenyong Zhu | Guwei Sun | Jingzhou Yuan

With the rapid development of life sciences at the end of the last century, the completion of the Human Genome Project (HGP) and the implementation of the follow-up functional genomics program, drug research and medical drug industry have entered a new historical period. As we all know, the development of new drugs is a high-input, inefficient work. With the rapid development of computer chemistry and biology in the 21st century, virtual drug screening technology has become an important branch of drug chemistry development. However, the establishment of a reasonable pharmacophore model to accurately determine or predict the molecular structure of the target protein, accurate and rapid calculation of candidate compounds and target interaction between the free energy change is the key to virtual drug screening, but also to limit the accuracy of virtual screening bottleneck. Virtual screening techniques can be divided into two types based on the receptor biomacromolecule structure and the ligand-based small molecule. Virtual screening often requires good predictive performance in practical drug design commercial applications. There are a large number of virtual screening methods based on machine learning, which are used to extract the characteristics of compound molecules, such as various molecular fingerprints and molecular descriptors, and use Bayesian statistical methods, nearest neighbor methods, support vector machines and artificial neural Network and other machine learning methods for compound molecules for high-throughput virtual screening. In particular, in recent years, in-depth learning methods have been successfully applied to the study of ligand-based drug virtual screening and molecular fingerprint generation, and show better predictive performance. Based on the virtual learning method of machine learning, the general practice is to use software to obtain fixed-length compound characteristics, including molecular fingerprints and molecular descriptors, and then call the machine learning method to build the model. However, the manual extraction of these general methods is usually associated with a semantics that is immutable and independent of the corresponding drug target and its activity activity, which can not compensate for the characteristics of manual extraction and the activity values ​​acting on different drug targets The gap. At present, molecular fingerprints with good properties are usually larger in virtual screening based on ligand activity. The Therefore, in today's actual virtual drug screening, an urgent need to be able to get relatively short, better performance of the molecular fingerprint generation method