AI for drug development
Russian scientists from Insilico Medicine, using a new artificial intelligence system called Generative Tensorial Reinforcement Learning (GENTRL), discovered 6 promising methods of treating fibrosis in 21 days, and the total time for modeling, synthesis and validation of molecules was 46 days.
Fibrosis is an overgrowth of connective tissue with the formation of scars as a result of chronic inflammation. Such isolation leads to a gradual loss of specific functions of the affected organ (for example, pulmonary insufficiency in pulmonary fibrosis). One of the causes of fibrosis is the activity of the enzyme DDR1, and accordingly, a possible way to stop the development of the disease may be a decrease in the activity of this enzyme.
GENTRL has generated 6 variants of drug molecules capable of coping with the task. One of the proposed variants was tested on mice and showed favorable results.
The system is based on the algorithm of the so-called generative-adversarial network, built on a combination of two neural networks, one of which (network G, generative) generates samples, and the other (network D, discriminative) tries to distinguish genuine samples from incorrect ones, i.e. networks G and D have opposite goals – to create samples and discard unnecessary ones. The use of this technique allows, in particular, to generate photographs that are perceived by the human eye as natural images. In addition, networks can be used to improve the quality of fuzzy or partially corrupted photos, as well as to obtain new molecular structures with specified properties.
In order to create GENTRL and thereby significantly speed up and improve the process of obtaining new drugs, scientists have been developing a theoretical basis for generative-adversarial networks and other similar machine learning methods for more than three years.
Drug development and validation is a rather time–consuming process that takes a lot of time and rarely reaches human clinical trials. Even a slight acceleration of the process will lead to significant savings and public benefit.
The article by Zhavoronkov et. al Deep learning enables rapid identification of potential DDR1 kinase inhibitors is published in the journal Nature Biotechnology.
Elena Panasyuk, portal "Eternal youth" http://vechnayamolodost.ru based on Eurekalert: Novel molecules designed by artificial intelligence in 21 days are validated in mice.