— Sechenov University is now opening Russia’s first center for artificial intelligence in pharmaceuticals. What tasks do you set for yourself?
— When talking about using artificial intelligence methods in drug development, it is generally meant that the neural network will simulate the molecules necessary to create a new drug. However, in this case, a new drug will only be able to be created in 10-15 years when all the necessary tests on cell and animal models are carried out, the dosage form is developed and clinical studies are conducted. Such deadlines are associated with a large number of different tests that must be carried out at all levels to prove the effectiveness and safety of a new drug.
At the same time, there is another very important task in drug development – the development of technologies for the production of the dosage form itself, for example, tablets or an injection solution.
This process is quite complex technologically, but the number of parameters that affect the result is quite clear and this process can be digitized much faster. Therefore, we will use artificial intelligence not to create new molecules, but to accelerate the development of finished dosage forms of generic and original drugs.
But we will start with generic drugs, which are the most common drugs in the Russian and international markets. Moreover, the creation of artificial intelligence-based tools in this field does not require changes in the regulatory field that regulates the drug development process extremely tightly.
— A generic is a copy of a drug. They are produced after the patent expires and the monopoly of the patent owner company. However, the patent discloses the composition of the drug. Still, why is it difficult to produce and why is artificial intelligence needed here?
— You can make an analogy with cooking: You want to cook a meal, ask your friend for the recipe and cook accordingly. But it turned out to be tasteless, everything is correct, although it seems that the ingredients are at hand. Caused? Yes, because if you ask the stewardess in more detail, it turns out that she did not mention many nuances.
In fact, it turns out that, for example, when adding eggs, the housewife separates the yolk from the white and does not combine them. All this was placed on the stove with the addition of oil at a certain temperature, for example, at first for 20 minutes, not 35. These nuances need to be known and are not specified in the patent.
Technological nuances are one of the biggest secrets in the production of any medicine, be it generic or original.
So, we have the composition of the drug, we know which molecule was used, but taking a tablet whose properties are exactly the same, it does not matter how and where it should be broken down, how long it will take to release the active ingredient. easiest task.
– What do they do in these situations?
— Technologists are examining regulatory documents, patents, open sources and trying to reproduce the technology to create the drug. Sometimes it’s easier to do that, and sometimes the technologies aren’t actually defined at all, you have to figure it out. Then you need to scale – which is also a problem, as there are production nuances here. Making medicine in a laboratory is one thing, making it in large quantities in a factory is another.
We plan to create and train a neural network model to help develop this technology and then scale it. This will make it possible to produce the finished dosage form faster and with better quality, whether it is the generic or the original drug that is currently needed to provide import substitution and medicine to our patients.
— So you are working on an artificial intelligence that will create this drug production technology for a technologist?
“We are working on tools based on artificial intelligence methods that will help technologists speed up the process.
Let’s take a general example. We ask a technologist to recreate the production of the drug: He sits down for a month or two, collects the necessary information, then presents technology options for creating the drug. Of these, the one closest to the production site is selected. At the same time, raw materials and materials are selected; these also affect technology.
The technologist then manually creates several different tablets with different technological approaches and sends them to the analytical laboratory for examination. There analysts look at which tablet is most similar to the original. If the point cannot be reached, the process is repeated. It takes an average of 12-15 months.
If we are talking about artificial intelligence, it significantly speeds up the process of collecting information, eliminates everything unnecessary, minimizes the number of errors and provides the most rational technological options: how to make technology, what raw materials to make it from, in what formats and with what to work.
— Have these AI technologies already been created?
– We just have begun. Our team has extensive experience in drug development: we have produced more than 200 generic drugs. Our colleagues at the Institute of Systems Programming of the Russian Academy of Sciences, led by Academician Harutyun Avetisyan, have extensive experience in creating solutions in the field of artificial intelligence.
We want to combine these competencies because now all Russian pharmaceuticals deal with generic products. Therefore, the task of accelerating their production is demanded now, not after 5-7 years.
— What are your next steps in applying AI technologies?
— We will train the neural network based on our experience in producing drugs. Unfortunately, we cannot use the data resulting from the creation of these 200 drugs because each development is conducted as a trade secret.
That’s why we need to collect data in collaboration with industrial partners. To do this, the technologist, together with the IT specialist, must go the entire path of creating a medicine, not just a drug. There are many tricks here because the neural network needs to take into account the technical and economic parameters of a particular production. For example, what are the conditions in this production, what raw materials are available.
— How much data do you plan to train the neural network model on?
— This is a trade secret related to our agreements with our industrial partners.
— Approximately when will it be ready?
— The work program has been designed for 3 years with different solutions that will be economically feasible. Nobody needs a RUB 20 million solution that will only reduce production by two weeks. For this amount, you can hire a few more tech experts and somehow speed up or parallelize the process.
We will be able to reveal the technological results of our work within one and a half to two years. During this period, we will develop drugs and collect data.
— When the neural network is ready, how much will it shorten the drug development time compared to your expectations?
— If we talk about generics, we hope for 30-40%. Now the process takes about 15 months, it would be great if we can reduce it to 10 months.
Plus, registration also takes about a year and a half, taking into account the study of the drug’s bioequivalence in patients or healthy volunteers, depending on the type of drug. The Ministry of Health will receive a high-quality dossier, which we hope will contain a minimum of questions during the state examination. This shortens the time it takes to bring the drug to market, and reducing this time by 3-6 months means saving many people’s lives and protecting health.
There are definitely some stages that will be reduced: data collection and analysis. There are also things to do with the accumulation of sample numbers. Here we cannot say whether the neural network will help reduce their number, by how much, and how these changes will affect the overall timing of drug release.
If we imagine, the ideal would be to shorten the process of creating a generic and obtaining the registration file by up to one year. Then patients and the country as a whole will be protected from a significant number of risks. Additionally, this will add competitiveness to our pharmaceutical industry.
— Is there any example of this in the world?
— Currently, there is no industrially applied solution in this field with proven economic impact. I don’t think anyone does this. They just, as a rule, do not shout about such things, they are gradually introduced into their production.
I repeat: Production technologies are the strictest trade secret, a regime of secrecy. Whoever produces new or copied drugs faster and cheaper will gain a good advantage.
– It seems that the time will come when it will be possible to make any generic drugs in Russia within a year, without waiting for drugs from the West?
— Yes, under current conditions, this will be a serious success and a very useful development.