— Sechenov University is inaugurating Russia’s first center dedicated to artificial intelligence in pharmaceuticals. What ambitions guide this move?
— In discussions about using artificial intelligence to accelerate drug development, the focus often centers on simulating molecules to craft a new medicine. Yet the creation of a truly new drug typically spans 10 to 15 years, as extensive testing on cells and animals, formulation development, and comprehensive clinical trials unfold. These timelines reflect the multitude of checks needed to prove safety and efficacy at every stage.
At the same time, a crucial objective in drug development is to refine the technology used to produce the dosage form itself, whether tablets or injectable solutions.
This pathway is technologically intricate, yet the number of influential parameters is finite, and the process can be digitized more rapidly. Consequently, artificial intelligence will be leveraged not to invent new molecules but to speed up the preparation of finished dosage forms for both generic and original medicines.
The starting point will be generic drugs, which are the most commonly used medications in Russia and on international markets. Importantly, the introduction of AI tools in this area does not require changes to the already stringent regulatory framework governing drug development.
— A generic is a copy of a drug. They come to market after patent expiration and the patent owner’s monopoly ends. The patent reveals the drug’s composition. Yet why does manufacturing generics remain challenging and why is AI needed here?
— It can be likened to cooking: you ask for the recipe and follow it, yet the result may fall flat. The reason often lies in details not fully captured by the recipe.
For instance, when adding eggs, a home cook might separate the yolk from the white and not combine them in the same way. The mixture is heated at a precise temperature for a specific duration. Such nuances are not always spelled out in patents.
Technological subtleties are among the most closely guarded secrets in the production of any medicine, whether generic or original.
So, even with the drug’s composition known and the active molecule identified, producing a tablet with identical properties involves more than the breakdown site and release timing of the active ingredient. That remains the easier part.
– What happens in practice?
— Technologists study regulatory documents, patents, and open sources to reproduce the technology and create the drug. Sometimes the path is straightforward; other times the methods are not defined at all and must be inferred. Scaling up is another hurdle, as laboratory production differs from factory manufacturing.
The plan is to build and train a neural network model to assist in developing this technology and then scaling it up. This aims to produce finished dosage forms faster and with higher quality, whether for generics or original medicines that are essential for import substitution and patient access.
— So the goal is to craft AI that helps generate drug manufacturing technology for technologists?
“The focus is on AI tools that empower technologists to accelerate the process.”
Consider a general scenario. A technologist attempts to replicate drug production. He spends weeks gathering data, then presents several technology options. The choice often lands on the approach closest to the production site. Raw materials and components also influence the technology.
The technologist then manually produces several tablets using different manufacturing approaches and sends them to an analytical lab for evaluation. Analysts determine which tablet most closely matches the original. If the target falls short, the cycle starts anew, typically taking 12 to 15 months on average.
With artificial intelligence, data gathering is accelerated, extraneous steps are eliminated, errors are minimized, and the most rational options emerge: how to design the process, which raw materials to use, in what formats, and with what equipment.
— Have these AI approaches already been implemented?
— The work has just begun. The team brings extensive experience in drug development, having produced more than 200 generic drugs. Partnerships with the Institute of Systems Programming of the Russian Academy of Sciences, led by Academician Harutyun Avetisyan, bring deep expertise in AI solutions.
There is a desire to fuse these competencies because most Russian pharmaceuticals are generic today. The imperative is to accelerate their production now, not five to seven years from now.
— What are the next steps for applying AI technologies?
— The neural network will be trained using internal experience in drug production. Data from the development of those 200 drugs cannot be reused since each project is treated as confidential information. The plan requires data-sharing with industrial partners, ensuring the technologist and IT specialist follow the entire medicine development path, not just the drug itself. The neural network must account for the technical and economic parameters of a given production, including production conditions and available raw materials.
— How much data will be used to train the model?
— That detail remains a trade secret tied to industrial agreements.
— When might results be visible?
— The work plan spans three years, with multiple viable solutions. No one needs a multi-million ruble solution that only shortens production by two weeks. The focus is on economical improvements that can be implemented widely. Results in technology development and data accumulation should emerge within one and a half to two years.
— How much time could a neural network save in drug development?
— For generics, a 30 to 40 percent reduction is anticipated. The current timeline of about 15 months could potentially compress to around 10 months. Regulatory procedures add roughly a year and a half, depending on bioequivalence studies. A robust dossier submitted to the health ministry should speed up state examinations, and shaving 3 to 6 months from the timeline could save lives and improve public health.
There will be reductions in data gathering and analysis time, and improvements in sample numbers. It remains to be seen how much the neural network will influence these aspects and the overall timing of market release.
If all goes well, the goal is to shorten the generic development and registration process by as much as a year. That would shield patients and the country from more risk and boost the competitiveness of the pharmaceutical sector.
— Is there a global precedent?
— Currently, there is no widely adopted industrial solution in this area with proven economic impact. Many players avoid public discussion and gradually integrate advances into production.
It is clear that manufacturing technologies remain highly guarded. Whoever can deliver faster and cheaper drug production gains a strong advantage.
— Could a scenario arise where Russia produces a majority of its generics within a year, without waiting on Western supplies?
— Under present conditions, such a development would be a notable achievement and a valuable step forward for public health and industry resilience [citation: internal project briefing].