0
0
A groundbreaking study published in Nature unveils a pivotal discovery by MIT researchers utilizing deep learning to uncover a class of compounds with the potential to combat drug-resistant bacteria, specifically targeting methicillin-resistant Staphylococcus aureus (MRSA). MRSA causes over 10,000 fatalities annually in the United States and poses a significant health threat due to its resistance to conventional antibiotics.
The compounds identified in the study demonstrated efficacy in eradicating MRSA in laboratory conditions and within two mouse models infected with MRSA, showcasing promising results. Remarkably, these compounds exhibited minimal toxicity against human cells, marking them as prime candidates for drug development.
A significant breakthrough of this research lies in understanding how the deep-learning model predicted antibiotic potency. Traditionally, these models operate as “black boxes,” leaving scientists unaware of the criteria underlying their predictions. MIT’s study aimed to unravel this mystery, providing insights into the model’s decision-making process, thereby paving the way for designing more effective antibiotics.
Professor James Collins, leading the Antibiotics-AI Project at MIT, highlighted the project’s mission to unearth new antibiotic classes targeting seven deadly bacteria strains within seven years. This latest achievement represents a critical milestone in this endeavor.
Utilizing an expanded dataset comprising approximately 39,000 compounds tested for antibacterial activity against MRSA, MIT researchers trained a deep learning model to recognize chemical structures associated with antimicrobial potential. This enabled the model to predict the likelihood of a compound possessing antibacterial properties based on its molecular structure.
To decipher the model’s predictions, an adapted algorithm called Monte Carlo tree search was employed, revealing not only the antimicrobial activity estimation but also identifying specific molecular substructures contributing to this activity.
By combining predictions of antimicrobial activity with toxicity assessments on human cells, the researchers sifted through 12 million compounds. The models pinpointed five different classes of compounds with predicted activity against MRSA. Following validation experiments, two promising compounds from the same class exhibited potent antibacterial properties in both laboratory dish tests and mouse models, significantly reducing MRSA populations.
These compounds disrupt bacterial cell membrane integrity by interfering with their electrochemical gradients, critical for essential cellular functions. Importantly, this disruption targets bacterial membranes selectively without substantial damage to human cell membranes.
Phare Bio, a nonprofit associated with the Antibiotics-AI Project, aims to delve deeper into these compounds’ clinical potential, while MIT continues its efforts to design additional antibiotics based on these findings. This breakthrough showcases the potential of advanced computational methods in revolutionizing drug discovery, offering hope in the fight against drug-resistant bacteria and bolstering efforts to develop more effective antibiotics.
