Insilico's Revolutionary Drug Discovery Engine

Drug discovery is a painstakingly long process. It begins with testing thousands of molecules to just find a few lead-like compounds which are likely to be useful. What's more, only about one in ten of those get passed clinical trials in human patients. As you can imagine, even a small decrease in the time it takes to discover new drugs is a profound improvement.

Hong Kong-based Insilico Medicine, a leader of artificial intelligence in drug discovery, recently published a paper titled "Deep learning enables rapid identification of potent DDR1 kinase inhibitors". Generative tensorial reinforcement learning (GENTRL) - the name for the system - proved capable of finding six treatments within 21 days. One lead was even proven favourable in mice.

The company was founded back in 2014 by Alex Zhavoronkov. After working at ATI for a few years, his interest shifted to biotechnology. Particularly, the possibility of extending the lifespan of a person through AI solutions. Over time, Zhavoronkov found an interest in generative adversarial networks (GAN), which was invented by Ian Goodfellow and his colleagues in 2014.

GENTRL is based on Insilico's groundbreaking research in 2016 about using GAN and generative reinforcement learning (RL) to accelerate drug discovery.

When we first proposed the idea of using the AI technique of generative adversarial networks to accelerate drug discovery in 2016, most of the industry was skeptical. With GENTRL’s successful experimentation and validation, Insilico has moved the use of AI for drug discovery from academic theory to reality, from what many thought was scientifically impossible to now possible.
— Zhavoronkov

Generative Adversarial Networks are a form of AI imagination. It involves two neural networks, one of which is called the generator. It works to produce new data instances from a given training set. Whereas the other neural network - the discriminator - evaluates them for authenticity by checking whether each instance of data it reviews belongs to the actual training dataset or not.

The research the company published based on this idea for drug discovery helped bring in investment money. Furthermore, since the publication, GANs were being explored for the accelerated generation of useful molecular structures. As such, Insilico paved a path for scientists worldwide to develop machine learning techniques to improve the drug discovery process.

Currently, the company has developed a comprehensive drug discovery machine, which utilized millions of samples and multiple data types to discover signatures of disease and identify the most promising targets for billions of molecules that already exist or can be generated with the desired set of parameters. This was demonstrated when GENTRL created six molecules, four of which could specifically inhibit discoidin domain receptor 1 (DDR1) activity - an enzyme involved in fibrosis. The time to generate the molecules was 21 days, and to validate them in vitro and in vivo took it up to just 46 days total. Moreover, on top of just taking a few weeks compared to the numerous years it takes to develop similar drugs, it also costs a fraction of the amount. Methods to develop drugs traditionally requires millions of dollars whereas Insilico's process was merely around $150,000.

Nevertheless, the drugs designed by GENTRL happened to be less effective than the already existing DDR1 kinase inhibitors developed by a team of scientists.

Their molecules are amazing, they’re a little bit better than what our AI could do. But again that’s years versus people who don’t have a lot of knowledge of chemistry doing this stuff.
— Zhavoronkov

Overall, Insilico's drug discovery tool is still developing but it shows a lot of hope and potential for quickening the process of drug discovery in addition to increasing the probability of success. The company is ushering in new possibilities for the creation and discovery of new life-saving medicine for incurable diseases.