CER's evolutionary biologists are helping the search for new antibiotics by testing artificial intelligence-based algorithms


In their latest research, evolutionary biologists at the ELKH Centre for Ecological Research (CER) have tested the operation of artificial intelligence-based algorithms to search for new molecules that could potentially be used as antibiotics. They have observed that these algorithms do not always work with equal efficiency, so the software used to identify each new antimicrobial molecule is of utmost importance. The paper summarizing the results of the study was published in the prestigious journal Scientific Reports.

Antibiotic-resistant bacteria are perhaps the greatest public health threat of our time, which is why the search for new antibiotics is one of the most important avenues of biomedical research. As the development of antibiotic resistance is an ongoing evolutionary process, an evolutionary approach to understanding and studying this problem is essential.

When a pathogenic bacterium is exposed to an antibiotic, it is usually unlikely to survive, as the active substance kills it in no time. However, bacteria reproduce very rapidly, mutations occur in their genes, and some of these mutations change the properties of the microorganism. These altered characteristics may include some that make the bacteria more resistant to antibiotics. As its counterparts die, the resistant bacterium suddenly gains a huge evolutionary advantage, as it can survive the onslaught of the antibiotic and the next generation will be almost exclusively its offspring.

This phenomenon is a well-known process of evolution that has been going on for billions of years. However, reckless antibiotic use in recent decades has accelerated the emergence and spread of increasingly drug-resistant strains of bacteria. For this reason, in addition to the responsible use of existing antibiotics, it is crucial that researchers continue to search for new antimicrobial molecules that can replace those that have become ineffective.

In nature, especially in animals and plants, there are countless antibiotic molecules, so there is no danger of them running out. The problem is rather that the discovery of new potential active substances by classical molecular biology methods is happening more slowly than resistant bacteria spread. Modern artificial intelligence algorithms can, however, help automate research, as the chemical structure of existing antibiotics can guide researchers to find which molecular features to look for in potential new drug molecules. However, these algorithms are not infallible, so testing their performance is a key priority.

"Based on the amino acid sequence of the proteins, these algorithms predict whether the protein may have antimicrobial efficacy," says Zoltán Rádai, a researcher at the Institute of Ecology and Botany at the ÖK, and lead author of the study. "The algorithms use state-of-the-art machine learning techniques, but they can only predict the probability of the unknown protein acting as an antibiotic based on the characteristics of the known antibiotics," he adds.

Algorithms play an indispensable role in modern antibiotic research, as they can help identify potential candidates before an expensive and cumbersome laboratory phase, speeding up the process significantly. Testing of the programs is essential to determine how effectively they can identify molecules from different sources that can be used as antibiotics.

Over 1,500 peptides (shorter amino acid chains than proteins) with known antimicrobial efficacy and 3,000 peptides with no (or no known) antimicrobial efficacy were used to test 20 different algorithms. The aim was to determine whether this software could separate some antimicrobial peptides from others. Tests were also carried out on whether different algorithms work equally well when using molecules from different sources.

"The result was that the various algorithms differed not only in their average success rate, but also in how well they recognized antimicrobial peptides isolated from different groups of animals," the evolutionary biologist added.  "This will allow researchers to select the software that can provide the most reliable results for the molecule they are studying."

Researchers around the world are extracting a large number of peptides from a wide variety of organisms. Some of this research is aimed at finding new antibiotics, while other studies are designed to answer fundamental research questions. The importance of this research in the fight against highly resistant bacteria, both now and in the future, is enormous. The importance of an evolutionary approach is illustrated by the significant role that disciplines traditionally far from medicine, such as ecology and evolutionary biology, can and will play in solving future health crises.