AI against antibiotic resistance
Artificial intelligence has helped develop new drugs against antibiotic-resistant bacteria
A source: RNF Press Service
Using machine learning methods, Russian scientists have managed to develop a number of peptide compounds with pronounced antibacterial activity. The data obtained will make it possible to create medicines for the treatment of severe diseases caused by antibiotic-resistant infections. The results of the study, conducted with the support of the Presidential Program of the Russian Science Foundation (RNF), are published in the journal Antibiotics (Bolatchiev et al., Novel Antimicrobial Peptides Designed Using a Recurrent Neural Network To Reduce Mortality in Experimental Sepsis).
In recent decades, humanity has faced a global problem of resistance of pathogens to antimicrobial drugs. The COVID-19 pandemic is also contributing. According to recent studies, no more than 7% of patients with coronavirus have a bacterial co-infection and, accordingly, need antibiotics, but in fact more than 70% receive these drugs. Often such an appointment is a reinsurance when the patient's condition is severe and there is no time to conduct long microbiological studies. As a result of the widespread use of antibiotics, not only in medicine, but also in agriculture, natural selection is triggered: microorganisms that have been able to develop protective mechanisms against drugs survive.
It is difficult to adjust the dynamics of antibiotic use on a global scale, so the most effective solution to the problem is the development of new drugs. This is a very expensive enterprise, which may not pay off because of the same resistance. Therefore, it is necessary to look for compounds to which bacteria do not develop resistance or form it no faster than in 20-30 years.
The question can also be approached from the side of evolution: for millions of years, living organisms have been in contact with pathogens and have learned to resist them. One of the oldest mechanisms of innate immunity is based on antimicrobial peptides, of which more than three thousand are now known. These molecules are able to fight pathogens of different classes and even tumors. In the case of bacteria, unlike antibiotics, they do not interact with any protein target whose genes can mutate, but directly destroy the cell wall. Such systems can be combined with antibiotics and increase the effectiveness of therapy due to the synergistic effect.
"Microorganisms have no protection against antimicrobial peptides, which means that they are a promising basis for a new generation of drugs. At the same time, it is expensive to synthesize them, and they are also quickly destroyed by the body's enzymes. We decided to try to create completely new peptides similar to natural ones, but more stable and, possibly, more effective," says Albert Bolatchiev, author of the work, project manager for the RNF grant, Candidate of Medical Sciences, employee of Stavropol State Medical University.
The staff of Stavropol State Medical University (Stavropol) used neural networks and artificial intelligence to develop a synthetic one based on natural antimicrobial peptides. These molecules represent an amino acid chain — the authors have "read" more than 3,000 such chains — and then revealed patterns in them. Based on the latter, the neural network generated 200 new unique peptides, but it is too difficult and expensive to synthesize all of them for the experiment, so the authors continued computer calculations.
At the next stage, they studied the sequences found using bioinformatic screening programs. These systems predict the potential activity of peptides against various microorganisms by analyzing the physicochemical and other properties of the molecules. Only five antimicrobial peptides passed such a rigorous selection, and they were tested on cultures of super-bacteria — antibiotic-resistant strains of Klebsiella and Pseudomonas aeruginosa, which cause a variety of infections, including nosocomial ones.
The results of the "in vitro" studies showed that out of five, only two compounds caused the complete death of microorganisms. They and another of the "non-working" peptides were tested on mice infected with a lethal dose of klebsiella, there was also a control group without treatment. In the latter, all the animals died on the third day, but the synthesized molecules showed unexpected results. Only one of the peptides that showed activity in culture worked on mice, increasing their survival rate to 50%, but completely "non—working" at all - up to 70%.
Probably, at the heart of this phenomenon are still unknown interactions with other molecules and pathways in the body that have yet to be revealed.
"For me personally, the most amazing thing about our results is that science has reached such a level of progress: artificial intelligence can create new pharmacologically active compounds! In our case, the neural network generated unique substances with pronounced antibacterial activity. We believe that this method can be applied to the development of other compounds of a peptide nature, and therefore we have set ourselves an ambitious task — to commercialize the technology. We will apply to state funds, for example, to Skolkovo, in order to attract investments to launch a startup. I am sure that right now is the right time for this," Albert Bolatchiev notes.
Portal "Eternal youth" http://vechnayamolodost.ru