Revolutionizing Antibiotic Discovery: AI's Promising Future
Written on
Chapter 1: The Dark Days of Infectious Diseases
Consider a bright day in a picturesque park. A cheerful child is playing football when suddenly, an opposing player tackles him, resulting in a painful scratch. This minor injury spirals into a severe infection, leading to tragic consequences.
This scenario, while unsettling, was a common reality prior to the 20th century. Infectious diseases proliferated, claiming lives even from minor injuries. Life expectancy hovered around 47 years, even in more developed regions.
The turning point came in 1928. Dr. Alexander Fleming, a Scottish physician and microbiologist, returned from vacation to find his lab in disarray. Among the chaos, he noticed that some Petri dishes with staphylococcal cultures had developed mold that inhibited bacterial growth.
Staring through his microscope, he realized he had stumbled upon Penicillium notatum, a mold producing a substance now known as penicillin—the first antibiotic. As Dr. Fleming reflected, “I certainly didn’t plan to revolutionize all medicine, but I guess that was exactly what I did.”
The period from the late 1940s to the 1960s is often referred to as the "golden era" of antibiotic discovery. Many essential antibiotic classes emerged during this time, such as β-Lactams, Aminoglycosides, and Tetracyclines.
With antibiotics, the world saw a decline in deaths from infectious diseases.
What Went Awry with Modern Antibiotics?
For those living today, life without antibiotics is hard to fathom. Unfortunately, we are racing toward a scenario reminiscent of the pre-antibiotic era. Current antibiotics are losing their effectiveness against evolving bacteria, fueled by natural mutations and the overuse of these medications.
The World Health Organization (WHO) has identified three particularly dangerous resistant bacteria:
- Carbapenem-resistant Pseudomonas aeruginosa
- Carbapenem-resistant, ESBL-producing Enterobacteriaceae
- Carbapenem-resistant Acinetobacter baumannii (CRAB)
These pathogens pose serious risks, causing infections in various body systems and resisting even our strongest antibiotics. Alarmingly, projections indicate that resistant infections could lead to 10 million deaths annually by 2050!
Why Aren't We Developing More Antibiotics?
Creating new antibiotics is a daunting challenge. The development process can span 10 to 15 years and cost over $1 billion. Each candidate must navigate extensive clinical trials, and disappointing results are common, discouraging pharmaceutical companies from pursuing new antibiotics.
In the past two decades, only a few new antibiotic classes have emerged, highlighting a concerning trend. According to a 2021 WHO report, only 27 antibiotics are currently in development, with minimal innovation.
"Time is running out for us to bring new antibiotics to market," warns Dr. Valeria Gigante of WHO's Antimicrobial Resistance Division. “Without immediate action, we risk reverting to a pre-antibiotic era where common infections could become fatal."
AI Enters the Scene
A groundbreaking study from MIT in 2020 showcased how deep learning could lead to the discovery of a new antibiotic named Halicin. Remarkably, Halicin was not structurally similar to existing antibiotics and had initially been investigated for diabetes treatment before being discontinued.
However, in the context of antibiotic use, Halicin proved highly effective against formidable pathogens like Mycobacterium tuberculosis and Clostridioides difficile.
In 2023, researchers published findings in Nature Chemical Biology, unveiling another antibiotic called Abaucin, effective against Acinetobacter baumannii without harming beneficial gut bacteria.
This marks a significant breakthrough—two novel antibiotics identified in just three years, a stark contrast to the decades-long drought in antibiotic discovery.
AI's Role in the Antibiotic Discovery Process
Historically, antibiotics were discovered by examining soil microbes for antibacterial properties. This method was effective for many years, but as researchers increasingly found similar compounds, it became clear that new strategies were needed.
Researchers turned to vast libraries of chemical compounds, yet screening millions of chemicals for antibiotic activity proved cumbersome. AI, particularly the Directed-Message Passing Graph Neural Network (D-MPGNN), has addressed this issue effectively.
What is a D-MPGNN?
D-MPGNNs are advanced graph neural networks designed to learn from data structured as graphs. In this context, chemical compounds are represented as graphs, with atoms as nodes and chemical bonds as edges.
By iteratively updating representations of nodes through a process called message passing, D-MPGNNs can capture both local and global structural information about a molecule. After processing, a pooling function aggregates these representations into a single fixed-size output that indicates the compound's potential as an antibiotic.
The Discovery of Halicin
Researchers screened 2,335 unique compounds for their ability to inhibit E. coli growth. Using a binary classification model called Chemprop, they identified Halicin as a promising candidate, demonstrating strong broad-spectrum bactericidal activity.
Chemical Structure of Halicin
The Discovery of Abaucin
Similarly, the team screened 7,684 molecules against Acinetobacter baumannii. The D-MPGNN model identified RS102895 as a potent inhibitor, which was later renamed Abaucin. This antibiotic selectively targets harmful bacteria while preserving beneficial gut flora.
Chemical Structure of RS102895 (Abaucin)
AI's influence on antibiotic discovery is remarkable. What are some of the most notable AI achievements you've encountered? Share your thoughts in the comments!
Loved This Story? Subscribe to my Substack newsletters below!