In silico

Japanese pharmaceutical company starts testing a drug synthesized using a neural network

ragequit, Habr

Pharmacists and programmers from Sumitomo Dainippon Pharma and Exscientia companies threw firewood into the fire of the dispute "where the independence of the machine should end and manual control should begin."

On January 30, a press release was published on the official website of Exscientia, stating that with the help of the AI developed by them, Sumitomo Dainippon Pharma has determined the formula and synthesized the active substance for a drug against obsessive-compulsive disorder (OCD).

Of course, Exscientia marketers call their development "Artificial Intelligence", but the development of a new drug was carried out with the help of a learning neural network. In fact, the neural network has determined the formula of a new drug through the search and analysis of combinations of known active substances. Now the development and synthesis have been completed, and SDP, the drug manufacturing company, is moving to the first phase of clinical trials on animals.

One of the advantages of involving neural networks to determine the formula of the active substance of medicines is the speed of the machine: the network, just like scientists, sorts out possible combinations of substances and builds a forecast of their main effect, however, unlike a team of biochemists, neural networks take much less time to create draft formulas. So, SDP specialists, with the technical support of Exscientia, coped in 12 months instead of the usual ~ 4.5 years that such studies usually last.

The developed substance was named DSP-1181, and its scope is the fight against OCD by increasing the response of the serotonin 5-HT1A receptor, that is, the active substance is a receptor agonist. The researchers claim that DSP-1181 is one of the most promising drugs for the treatment of OCD that currently exist.

The representative of Sumitomo Dainippon Pharma, represented by the company's senior executive director Toru Kimura, says about the latest development: "We are very excited about the results of a joint study that allowed us to develop a candidate compound in such a short time. Our experience in the search for new drugs based on GPCR monoamines, together with the capabilities of AI, allowed us to work effectively and provided a successful result. We will continue to work hard to ensure that our medicine passes clinical trials and gets to patients as soon as possible."

Unfortunately, the representatives of Exscientia did not provide any technical details of the development in the press release, but the following picture can be drawn from fragments of individual phrases:

• a neural network was used;
• The Japanese company SDP has provided an array of labeled data for training and networking, as they have "experience in finding a drug based on GPCR monoamines (monoamine neurotransmitters)";
• the neural network was engaged in mechanical sorting of candidate compounds for the active substance, after which the generated data array was checked by scientists of the pharmaceutical company;
• there is nothing breakthrough in the technology itself.

It is worth noting that for all the banality of the process of searching for the active substance in the described case (and filtering a large array of data using machine learning is now used in many fields, from mathematics and astrophysics, to oil production and geological exploration), the very fact that the formula of the active substance of the drug for such an unpleasant neurological disease as OCD was originally derived by a machine – impressive. The most important thing that Sumitomo Dainippon Pharma and Exscientia have done is to expand the horizon of the application of modern computing machines and capacities in the field of pharmaceuticals. But the development of a drug is not a fast and extremely expensive process.

According to Alexander Zhavoronkov, neural networks are already actively used in pharmaceuticals, but usually – to analyze existing scientific papers and the effectiveness of certain studies: has the drug gone on free sale, or has its development stopped on the basis of primary publication and has not passed clinical trials? Neural networks also processed clinical data of patients, and it is known about the direct involvement of machine learning in the search for an active substance with subsequent synthesis. Basically, the role of "AI" is reduced to processing an array of indirect data and preparatory work – in which direction researchers should move.

It is likely that with successful clinical trials, Sumitomo Dainippon Pharma and Exscientia will become one of the first companies that used a neural network and so quickly received the active substance of the drug for a wide range of consumers.

Portal "Eternal youth" http://vechnayamolodost.ru