New Machine Learning Model Finds Promising Antibiotics
Researchers from the University of Bern in Switzerland have developed a new machine learning model that can identify potential antibiotic candidates. This groundbreaking model focuses on ruthenium-based compounds, which have shown significant antimicrobial activity against resistant bacteria.
The growing problem of bacterial resistance to antibiotics poses a serious threat to modern medicine, as many hospital treatments rely on these drugs to control infections. Metal-based antimicrobials, including ruthenium complexes, have emerged as a potential solution. Compared to traditional organic carbon-based chemicals, metal compounds are ten times more likely to be effective against bacteria and are not necessarily more toxic to humans.
The team, led by Angelo Frei, utilized a combinatorial chemistry approach to create a library of 288 ruthenium compounds. They tested these compounds against methicillin-resistant Staphylococcus aureus (MRSA) and found a substantial percentage to be active. Using this data, they trained a machine learning algorithm to predict the activity against MRSA.
Taking it a step further, the researchers built a virtual library of 77 million ruthenium complexes and used the algorithm to identify two million potentially active structures. To validate the predictions, they tested a smaller sample of 54 structures in the lab against MRSA. Astonishingly, 53.7% of these compounds exhibited activity, representing a 5.7 times higher hit rate than the initial screening.
The unique properties of ruthenium complexes make them appealing candidates for drug development. These compounds have distinct mechanisms of action compared to traditional organic antibiotics, which could help overcome existing resistance mechanisms. Additionally, ruthenium complexes are known for their biocompatibility and low toxicity compared to other metal compounds. In fact, they are already being studied in clinical trials for treating cancer.
Nils Metzler-Nolte, an expert in bioinorganic chemistry, praised the versatility of this research method. He highlighted the vast number of compounds synthesized by the researchers, all with different three-dimensional shapes and properties. Organometallic complexes offer the potential for antibiotics with new and unprecedented modes of action.
While ruthenium is relatively expensive and scarce, the synthesis of these compounds is economical compared to commercially available drugs. Furthermore, the cost of drug discovery and development is primarily driven by clinical trials, rather than synthesis.
The next phase of research will involve experimental and computational validations to refine the predictions. Additional steps include further synthesis, characterization, biological tests, iterative design, and more. Molecular simulations will also play a crucial role in understanding the unique antibiotic mode of action exhibited by these metal complexes, as well as any potential resistance.
Lead author Angelo Frei emphasized the need to generate more data and larger libraries to cover a wider range of compounds in the periodic table. This will enable the model to predict more specific properties, such as the degree of activity and toxicity. Researchers hope that this innovative machine learning model can contribute to tackling the problem of antibiotic resistance and uncover new powerful antibiotics for medical use.
