Researchers using AI are mining the DNA of long-extinct species like woolly mammoths and giant sloths, looking for ancient genomic secrets to help fight today’s most infectious pathogens.
Every year, more than 1.25M people worldwide die from infections immune to drugs like antibiotics, according to the World Health Organization (WHO). By 2050, that number could rise to 10M people. And, within 6 years, around 24M people could fall into extreme poverty due to the high costs associated with treating infectious diseases, the WHO reports.
Dr. Cesar de la Fuente, a professor at the University of Pennsylvania, leads a team of researchers using AI to probe DNA from extinct, prehistoric creatures to identify novel solutions to dangerous drug-resistant microbes.
This technique, which Dr. de la Fuente and a team have dubbed “molecular de-extinction,” was detailed in a paper published in Nature Biomedical Engineering in June 2024.
“Exploring and comparing molecules throughout evolution can unlock new biological insights,” Dr. de la Fuente said. “Our AI-driven molecular de-extinction work lets us bring back molecules from the past that can help address contemporary challenges.”

Using a cluster of NVIDIA A100 GPUs, Dr. de la Fuente and his team at the University of Pennsylvania’s Machine Biology Group, trained a set of deep learning models to mine the proteomes of living and extinct species. The scientists hypothesized that today’s pathogens, which have spent decades adapting to and fighting modern-day drugs, might succumb to antimicrobial defenses hidden in the ancient genomes of extinct creatures.
The researchers trained 40 variants of deep learning models, named APEX, on DNA extracted from fossils of extinct bears and penguins, ancient Magnolia trees, straight-tusked elephants, sea cows, giant elk, woolly mammoths, and giant sloths.
The training used a combination of 988 peptides that the researchers created in-house, along with 5,093 and 5,500 publicly available antimicrobial peptides (AMPs) and non-AMPs, respectively. The researchers noted that the in-house dataset included 14,738 antimicrobial activity data values obtained from 34 bacterial strains.
The researchers ran the training using the cuDNN-accelerated PyTorch deep learning framework with a single NVIDIA A100 GPU.
After training, the models predicted encrypted peptide sequences—fragments of proteins that immune systems use to fight infections.
Specifically, APEX predicted more than 37,000 peptide sequences with broad antimicrobial applications; 11,000 were not found in living organisms—meaning their origin or the genomic inspiration originated in extinct creatures.
The researchers narrowed down their data and then synthesized 69 potential antibiotics from the APEX-generated peptides.
In lab tests with mice infected with a bacteria pathogen that commonly infects human burn victims, these ancient, resurrected peptides were able to fight off contemporary infections.
Researchers found that mice given an experimental antibiotic derived from giant sloths, named mylodonin-2, showed improved health within 2 days of starting treatment. In fact, mice treated with this sloth-derived drug improved at the same rate as mice in a control group treated with the common antibiotic, Polymyxin.
“Exploring extinct organisms allows us to access a vast array of molecules that contemporary pathogens have never encountered,” Dr. de la Fuente said. “Molecular de-extinction can provide a new arsenal of compounds to combat antimicrobial resistance, which is one of humanity’s greatest threats.”
In their paper, the researchers note that the de-extincted antimicrobial molecules attack microbes by depolarizing the inner membrane of a pathogen’s cells, which is different from most previously described antimicrobial peptides.
This novel recycling of DNA from extinct creatures is something even the researchers admit seems like it’s taken from a Michael Crighton thriller. It’s also a process that until recently, with the widespread use of GPUs, parallelized workloads, and AI, was almost impossible.

Dr. de la Fuente believes that generative AI holds the potential to revolutionize virtually every aspect of both traditional and novel drug discovery methods.
Developing a new antibacterial drug using traditional methods, including wetlabs and CPU-powered computers, can take 10 to 15 years and cost more than $1B. By using generative AI, Dr. de la Fuente says, both the cost and development times are significantly reduced.
“GPUs are transforming how we do our work in our lab,” Dr. de la Fuente said. “We can do in a few hours what it used to take 6 years of research to accomplish. This has enabled us to dramatically accelerate antibiotic discovery. It’s like bringing science fiction into reality.”
Dr. de la Fuente is in the early stages of setting up a company that can marketize the most promising antimicrobial drugs his research team discovers. The Machine Biology Group continues to use its APEX models to explore promising antimicrobial peptides that can help humans combat drug-resistant pathogens. The work is open source on GitHub.
Read the paper in Nature.
Review more research from Dr. de la Fuente’s lab.
The AI for Good blog series showcases AI’s transformative power in solving pressing global challenges. Learn how researchers and developers leverage groundbreaking technology and launch innovative projects using AI to create positive change for people and the planet.