More than 160 thousand new viruses were discovered using AI

An artificial intelligence system has just made the largest simultaneous virus discovery in history. In one test, the LucaProt deep learning algorithm discovered 161,979 novel RNA viruses and 180 viral supergroups hidden in 10,487 metatranscriptomes in bioinformatics databases around the world.

Viruses don’t usually show up when you glance under a microscope to see what’s new. The science uses genetic sequencing to find strands of DNA or RNA in a sample and then compares the results to previous databases. If the genetic sequence matches, it is a known virus. If it is not in the world records, then it is carefully examined to find a new microbe.

For their part, metatranscriptomes represent the collection of genetic information that a microbial community includes. They can be thought of as the sum of all the DNA or RNA data from a location, which is used to measure the level of complexity or diversity of an ecosystem. They may contain viruses that cannot be identified because the computers scanning the information simply do not find a match in the available databases.

But artificial intelligence (AI) can predict the structure of viruses and identify these patterns in large volumes of metatranscriptomic information. That’s exactly what a team of scientists from China and Australia did with LucaProt. They trained an artificial intelligence model capable of calculating the genetic variants of viruses and sent it to explore databases around the world. By improving the method, they discovered 161,000 new species of microbes in one movement.

“The vast majority of these viruses had already been sequenced and were in public databases, but they were so different that no one knew what they were. They made up what is often called the “dark matter” sequence. Our AI model was able to organize and classify all this disparate information, shedding light on the meaning of this dark matter for the first time,” explained Edwards Holmes, co-author of the study published in the journal cells.

Almost all living things are composed of deoxyribonucleic acid (DNA). But viruses are usually made from ribonucleic acid (RNA). It is a single-stranded structure instead of a double helix, and because it is less stable, it allows viruses to evolve quickly, resist their environment, and adapt to different hosts. Influenza viruses, Covid-19, HIV, Ebola, dengue or Zika are examples of RNA organisms.

“Previously, we relied on tedious bioinformatics pipelines to detect viruses, which limited the diversity we could explore. We now have a much more powerful AI-based model that provides exceptional sensitivity and specificity, allowing us to study viruses much more deeply. We plan to apply this model in several applications,” explained Sun Yat-se, another of the study’s authors.

The predictive potential of artificial intelligence systems used in science is bringing first results. The 2024 Nobel Prize in Physics was awarded to John J. Hopfield and Geoffrey E. Hinton, “the godfathers of artificial intelligence and machine learning” Their work underpins technology that helps predict proteins for future medicine. Meanwhile, hundreds of new figures were recently found in the Nazca Desert with the help of artificial intelligence. The archaeologists participating in the study considered the use of technology to be a new revolution in their field.