The Scientist Using AI to Hunt for Antibiotics in the Most Unlikely Places

The Human Body as Pharmacy

In a parallel line of research, de la Fuente has turned the AI's attention inward, examining the proteins and peptides produced by the human body itself. The human proteome contains thousands of proteins that serve a wide range of biological functions, from structural support to immune defense. By analyzing these proteins with machine learning models, his team has identified fragments that exhibit antimicrobial properties but had never been recognized as potential drug candidates.

This discovery has profound implications. If effective antibiotics can be derived from human proteins, they may offer advantages in terms of biocompatibility and reduced side effects. The immune system already uses antimicrobial peptides as part of its first-line defense against infection; de la Fuente's work suggests that the body may contain a much larger arsenal of antimicrobial compounds than previously appreciated, waiting to be identified and developed into therapeutic agents.

How the AI Works

The machine learning systems at the heart of de la Fuente's research operate by learning the relationship between a peptide's amino acid sequence and its antimicrobial activity. Trained on databases of known antimicrobial peptides and their properties, the models develop an understanding of the structural features that predict activity against different types of pathogens. They can then scan new sequences, whether from ancient genomes, human proteins, or environmental DNA, and assign a probability that each candidate will have useful antimicrobial properties.

The scale of this computational approach is what makes it transformative. Traditional antibiotic screening might evaluate a few thousand compounds in a year. De la Fuente's AI systems can analyze millions of candidate sequences in days, identifying hundreds of promising leads for laboratory testing. This dramatic acceleration of the discovery process is crucial given the urgency of the antimicrobial resistance crisis.

Once promising candidates are identified computationally, the team synthesizes them in the laboratory and tests them against panels of drug-resistant bacteria. The hit rate has been remarkably high compared to traditional screening methods, validating the AI's ability to identify genuine antimicrobial compounds from enormous datasets. Those that show activity in the lab then move through further testing to evaluate their safety and efficacy in animal models.

From Discovery to Impact

The challenge of translating computational discoveries into clinical treatments remains significant. Drug development is a long and expensive process, and the economic incentives that have driven pharmaceutical companies away from antibiotics remain largely unchanged. De la Fuente has been vocal about the need for new funding models, including government-backed pull incentives that guarantee a market for new antibiotics, to ensure that promising discoveries do not die in the laboratory.

Despite these challenges, the work represents a genuine reason for optimism in a field that has been defined by pessimism for decades. By demonstrating that AI can dramatically expand the universe of potential antibiotic compounds, de la Fuente has opened a door that other researchers are now walking through. Teams around the world are adopting similar computational approaches, creating a growing global effort that may finally begin to close the gap between the emergence of resistant infections and the development of new drugs to treat them.

The vision is ambitious but grounded in real results. The antibiotics of the future may come from the genomes of species that vanished millennia ago, from the proteins of our own bodies, or from the vast metagenomic databases that catalog the microbial diversity of every ecosystem on Earth. Thanks to artificial intelligence, we now have the tools to find them.

This article is based on reporting by MIT Technology Review. Read the original article.