Center for Strategic Assessment and forecasts

Autonomous non-profit organization

Home / Science and Society / New in Science / Other
Artificial intelligence will plunge into the universe of molecules in the search for amazing drugs
Material posted: Publication date: 05-12-2017
Dark night, away from city lights, the stars of the milky Way seem innumerable. But from any point visible to the naked eye not more than 4500 stars. In our galaxy, they are 100 to 400 billion galaxies in the Universe, and even more. Out in the night sky not as many stars. However, even this number gives us a deep story... of drugs and medicines. The fact that the number of possible organic compounds with medicinal abilities exceeds the number of stars in the Universe by more than 30 orders of magnitude. And chemical configurations that are created by scientists from existing medicines, akin to the stars that we could see in the center of the city at night.

Search of all possible drugs — an impossible task for humans, as the study of all physical space, and even if we could, the most part of discovered it would not fit our purposes. However, the idea that a miracle cure may be hiding amidst plenty, too tempting to ignore.

That is why we should use artificial intelligence that will be able to work more and to accelerate the opening. So says Alex Zhavoronkov, speaking at Exponential Medicine, San Diego last week. This application could be the largest AI in medicine.

Dogs, diagnosis and medication

Lark is the CEO of Insilico Medicine and CSO Biogerontology Research Foundation. Insilico is one of many startups developing AI that can accelerate the discovery of new drugs and medicines.

In recent years, said the lark, a well-known technique in machine learning — deep learning — made progress on several fronts. The algorithms are able to play video games — like poker player AlphaGo Zero or Carnegie Mellon — represent the largest object of interest. But the recognition of regularities — that gave a powerful impetus to deep learning, where machine learning algorithms finally began to distinguish cats from dogs and to do it quickly and accurately.

In medicine algorithms, deep learning, trained on databases of medical images, can detect life-threatening diseases with equal or greater accuracy than specialist human. There is even a possibility that the AI, if we learn to trust him, could be invaluable in the diagnosis of the disease. And like lark said, coming more applications and a track record will only grow.

"Tesla is already leading the cars on the street," says lark. "Three-and four-year technology already transports passengers from point A to point B at a speed of 200 kilometers an hour; one mistake and you're dead. But people trust their lives to this technology."

"Why not do the same in the pharmaceutical industry?".

Trial and error, again and again

In pharmaceutical research the AI will not have to drive the car. He will be an assistant that is paired with a chemist or two will be able to accelerate the discovery of new drugs, looking at more options in finding the best candidates.

Space for optimization and improving efficiency is just huge, says lark.

Search drugs — a painstaking and expensive business. Chemists screened tens of thousands of possible compounds in search of the most promising. Of these, only some go on to further study and even less will pass the test on the people, and of these all the crumbs are approved for further use.

This whole process can take many years and cost hundreds of millions of dollars.

This is the problem of big data (big data), and deep learning excels in working with big data. The first application showed that the system AI-based deep learning is able to find the barely visible patterns in huge data samples. Although drug manufacturers are already using software for screening compounds, such software requires clear rules, written by chemists. The advantages of AI in this case is its ability to learn and improve yourself.

"There are two strategies of innovation on the basis of the AI in the pharmaceutical industry, which will provide you the best molecules and the rapid approval," says lark. "One is looking for the needle in the haystack, and the other creates a new needle".

To find a needle in a haystack, the algorithms are trained on large database of molecules. Then they are looking for molecules with the appropriate properties. But to create a new needle? This opportunity is offered on generative adversarial network specializiruetsya larks.

These algorithms put two neural networks against each other. One generates a meaningful result, and the other decides whether the result is true or false, says lark. Collectively, these networks generate new objects such as text, image or, in this case, the molecular structure.

"We started to use this specific technology to deep neural networks imagined new molecules to make it perfect from the start. We need the perfect needle," says lark. "You may refer to this generative adversarial networks and ask her to create molecules that inhibit protein X at concentration Y, with the highest viability, desired characteristics and minimal side effects."

Lark believes that AI can find or produce more of the needles of the many molecular features, the release of the chemical to people so that they can focus on the synthesis of only the most promising. If it works as he hopes, we will be able to increase the number of hits, to minimize mistakes and to speed up the process.

The trick

Insilico not the only one engaged in the search for new ways to create medicines, and this is not a new area of interest. Last year, Harvard published a paper on the topic of AI, which is similarly selects candidates of the drugs. The software was trained on 250 000 drug molecules and used his experience to create new molecules that are mixed existing drugs and made suggestions based on desired properties. However, as noted by MIT Technology Review, the results are not necessarily meaningful or easily synthesized in a lab, and the quality of the results, as always, highly so, how is the quality of the data provided initially.

Stanford Chemistry Professor Vijay Pande said that the images, speech and text — which are currently interested in deep learning — a good and clean data. But chemical data, on the other hand, is still optimized for deep learning. In addition, although public databases exist, most of the data still lives behind closed doors of private companies.

To overcome all the obstacles the company Zhavoronkova focused on validating the technology. But this year, the skepticism in the pharmaceutical industry, it seems, is replaced by interest and investment. Even Google can break into the race.

As the evolution of the AI and hardware, the greatest potential is still to be discovered. Perhaps one day, all 1060 molecules in the field of drugs will be at our disposal.

Ilya Hel


RELATED MATERIALS: Science and Society