Scientists in the medical field are now using artificial intelligence (AI) technology to make new discoveries.
Recently, they have found the first new antibiotics in six decades, which is great news in the fight against drug-resistant Staphylococcus aureus (MRSA) bacteria. This type of bacteria is responsible for countless deaths worldwide every year, so this development brings hope for a brighter future.
A team of researchers, including James Collins, a professor of medical engineering and science at the prestigious Massachusetts Institute of Technology (MIT), recently unveiled a breakthrough.
By employing deep learning models that are more transparent, they successfully uncovered a new class of antibiotics capable of eliminating drug-resistant MRSA bacteria.
The unique aspect of this study was the researchers' ability to comprehend the reasoning implemented by the AI models when identifying potentially effective antibiotic molecules.
As stated by Prof. Collins, "Our work provides a time-efficient and resource-efficient framework, shedding light on chemical structures in ways that were unprecedented."
This development was published in the esteemed scientific journal Nature and involved collaboration among 21 dedicated scientists.
In parallel with these advancements, AI is now also significantly influencing the field of drug discovery itself.
AI tools such as life2vec, built on transformer models akin to powerful language-based tools like ChatGPT, have been designed to predict an array of critical information based on individuals' data, including health history, education, occupation, and income.
However, this tool, trained using extensive Danish population data, can accurately forecast lifespan and other crucial factors even better than established models.
Within the context of antibiotic research, the MIT team utilized a deep-learning model to forecast the activity and toxicity of the newly discovered antibiotic compound.
Deep learning, an AI technique that autonomously learns and represents data features without explicit programming, has gained momentum in drug discovery for accelerating candidate identification and property prediction.
For this specific research, the focus was on methicillin-resistant Staphylococcus aureus (MRSA), a highly threatening bacterium causing infections ranging from mild to life-threatening conditions.
In the European Union alone, nearly 150,000 MRSA infections arise every year, leading to approximately 35,000 annual deaths due to antimicrobial resistance, according to the European Centre for Disease Prevention and Control (ECDC).
The researchers used an expanded deep-learning model to develop and refine potential drug candidates. They evaluated around 39,000 compounds to determine their antibiotic activities against MRSA.
They also integrated toxicity evaluations on various human cell types using three separate deep-learning models. By merging antimicrobial activity and toxicity predictions, the team identified compounds that effectively combat MRSA while minimizing harm to human cells.
The screening process encompassed a pool of approximately 12 million commercially available compounds.
As a result, AI-driven models discerned five different classes of compounds demonstrating predicted activity against MRSA, each classified according to specific chemical substructures within the molecules.
While the predictive power of these models is evident, it's critical to highlight that they should not be employed for real-world predictions concerning individuals.
Tina Eliassi-Rad, a professor of computer science at Northeastern University, emphasized the need for caution, stating that the tool is specific to a particular data set and population.
In an effort to ensure responsible and human-centric AI development, the researchers actively engaged AI ethics experts and social scientists during the project.
By comprehending society's dynamics from a new perspective, these tools enable analysis of existing policies and regulations, providing valuable insights into ground reality.