Machine Learning Uncovers New Ways to Kill Bacteria With Non-Antibiotic Drugs
The discovery of antibiotics in 1928 forever transformed human health, providing a solution to lethal infectious diseases like pneumonia, tuberculosis, and sepsis. Antibiotics made surgeries safer, vastly reducing infection-related complications. However, antibiotics have their limitations, most notably the ability of bacteria to evolve resistance, which poses a significant global health threat. In 2019 alone, antibiotic-resistant bacteria, or “superbugs,” were responsible for an estimated 1.27 million deaths globally, a number expected to rise in the coming years.
In the fight against bacterial resistance, new scientific discoveries are offering promising alternatives. Research shows that nearly a quarter of drugs not traditionally used as antibiotics—such as cancer, diabetes, and depression medications—have the potential to kill bacteria at doses typically used in human treatments. This discovery raises important questions about how these non-antibiotic drugs can affect bacteria and whether they can help combat bacterial infections. If these drugs kill bacteria in ways that differ from known antibiotics, they may serve as vital leads in developing new types of antibiotics. On the other hand, if they act similarly, prolonged use could inadvertently promote antibiotic resistance, a significant concern in chronic disease treatments.
A research team from the Mitchell Lab at UMass Chan Medical School, along with other scientists worldwide, is leveraging machine learning to address the growing challenge of bacterial resistance. The team has used genetic screening to study the genetic mutations in bacteria that make them resistant or more sensitive to drugs. Their goal is to understand how non-antibiotic drugs kill bacteria, potentially opening new doors for antibiotic discovery.
Machine Learning Reveals New Killing Mechanisms
To better understand how non-antibiotic drugs can kill bacteria, the researchers employed a genetic screening method to investigate the mechanisms of anticancer drugs on bacterial cells. This technique enabled them to identify which specific genes and cellular processes were affected when bacteria mutated, providing insights into how these drugs kill bacteria. By analyzing millions of toxicity instances across 200 drugs and thousands of bacterial mutants, they grouped the drugs into networks based on their effects on bacteria.
Interestingly, antibiotics formed tight clusters based on their known killing mechanisms, such as targeting bacterial cell walls or interfering with DNA replication. However, when non-antibiotic drugs were added to the analysis, they formed distinct clusters, suggesting they target bacteria in novel ways. While the exact killing mechanisms for each drug remain elusive, this data grouping indicates that drugs within the same cluster likely work similarly. This finding demonstrates the potential of machine learning to uncover new bacterial targets for future antibiotics.
A crucial breakthrough came when Carmen Li, a colleague of the research team, sequenced the genomes of bacteria exposed to non-antibiotic drugs. The team pinpointed a bacterial protein targeted by triclabendazole, a drug used to treat parasitic infections. Notably, this protein is not targeted by existing antibiotics, highlighting its potential as a new bacterial target for antibiotic development. Moreover, two other non-antibiotic drugs were found to work similarly, demonstrating the power of machine learning to uncover previously unknown drug mechanisms.
Helping Antibiotic Discovery
The findings from this study provide researchers with multiple avenues for exploring how non-antibiotic drugs can work differently from traditional antibiotics. The combination of genetic screening and machine learning used in this research offers a solution to the critical bottleneck in antibiotic discovery—finding new ways to kill bacteria. Traditional antibiotic discovery methods involve screening thousands of chemicals and identifying how they kill bacteria, a process that is often resource-intensive and results in the identification of chemicals that work similarly to existing antibiotics.
Machine learning models provide an efficient alternative by identifying chemical compounds that target bacteria in novel ways. For example, MIT researchers recently used machine learning to develop a new antibiotic, halicin, which kills drug-resistant bacteria through a mechanism that bacteria find difficult to resist. Halicin disrupts the bacteria’s electrochemical gradient, a process essential for ATP production, thereby killing the bacterial cells. The researchers also found that bacteria did not develop resistance to halicin during a 30-day treatment period, in stark contrast to conventional antibiotics, which often encounter resistance within days. Halicin’s development showcases how machine learning can accelerate the discovery of antibiotics and identify molecules that are effective against drug-resistant bacteria.
Potential Impact on Global Health
The implications of this research are vast, especially as the world grapples with the rising threat of antibiotic-resistant bacteria. By identifying new bacterial targets and understanding how non-antibiotic drugs kill bacteria, scientists can develop novel antibiotics that bacteria are less likely to resist. Additionally, this approach may offer a solution to the slow pace of new antibiotic development, which has left many bacterial infections, such as methicillin-resistant Staphylococcus aureus (MRSA), increasingly difficult to treat.
Researchers plan to continue expanding their machine learning models, screening additional molecules to discover new antibacterial compounds. In one study, MIT researchers screened over 100 million molecules and identified 23 promising candidates, two of which demonstrated potent antibacterial activity in laboratory tests. This highlights the potential of machine learning to dramatically accelerate the antibiotic discovery process.
Machine learning’s ability to sift through vast chemical spaces efficiently offers new hope for combating the global crisis of antibiotic resistance. Researchers aim to use these models to not only discover new antibiotics but also optimize existing ones, making them more effective and safer for human use. This technology has the potential to revolutionize the way we approach bacterial infections, ensuring that we remain ahead in the ongoing battle against drug-resistant bacteria.
In conclusion, the integration of machine learning into antibiotic discovery is a promising development in the fight against bacterial resistance. By uncovering new ways to kill bacteria and identifying novel drug targets, researchers are paving the way for the next generation of antibiotics. As bacterial resistance continues to rise, these innovations are critical to safeguarding public health and ensuring that life-saving treatments remain effective for generations to come(UMassMed)(MIT for a Better World).
