New AI tool exposes plant pathogens and the genes behind them

Can a bacterium's DNA reveal whether it will make a plant sick? Bioinformatician Guillermo Guerrero Egido believes so. In his PhD research, he developed a computational method that not only identifies potential pathogens, but also the genes that may be responsible for their disease-causing abilities.

"There is now an enormous amount of bacterial DNA available in public databases," says Guerrero Egido. "We wanted to use those data to predict bacterial lifestyles. Does a bacterium live in soil, on plants or in the human body? Does it help plants grow, or does it make them sick?"

His new AI-based tool bacLIFE allows researchers to answer such questions much faster. This could help them identify beneficial bacteria for agriculture and spot potential pathogens at an earlier stage.

Guerrero Egido defended his thesis, "Bacterial lifestyle studies in the genomics era," on Nov. 11 at Leiden University.

Looking for genetic differences and similarities

For his research, Guerrero Egido analyzed the genomes from more than 16,000 bacteria belonging to two plant-associated bacterial groups. Scientific literature provided information on the lifestyles of around 40% of these bacteria, including whether they cause disease. These known examples were used to train the system.

"The basic question is: what do you have that I don't?" Guerrero Egido explains. "If one bacterium causes disease in plants, but its close relatives do not, which genes does it have that the others lack?" These genes could explain why this particular bacterium causes disease.

This approach, known as comparative genetics, has been used before. However, while researchers traditionally compare only a handful of bacteria at a time, bacLIFE can analyze thousands simultaneously. Using machine learning, a form of AI that learns from examples, the system identifies patterns associated with particular lifestyles.

The program identified hundreds of candidate genes that may be involved in causing plant disease. It can also make predictions about bacteria for which little information is available.

From computer to laboratory

Guerrero Egido wanted to know whether the tool's predictions were correct. To find out, he moved from the computer to the laboratory, where he investigated six candidate genes. The genes were switched off, after which he and his colleagues tested how well the bacteria could still infect plants.

That represented a new challenge for Guerrero Egido, whose work is normally primarily computational. "For this part of the project, I really needed help from my colleagues in the lab," he says.

The results were promising. Once a candidate gene had been disabled, the bacteria became far less effective at infecting plants.

Remarkably, some of the tested genes were still largely unknown. "We still do not know exactly what these genes do or how they work," says Guerrero Egido. "But we do know that they play a role in the bacterium's ability to cause disease."

Looking beyond plants

The method may also be useful beyond plant research. When analyzing data sets that included the bacterium responsible for tuberculosis, bacLIFE successfully identified several genes that were already known from previous studies. This suggests that the approach may also work for bacteria that infect humans or animals.

Guerrero Egido has also developed an adapted version for fungi, although this version still awaits experimental validation.

Narrowing down the search

According to Guerrero Egido, the new method could save researchers a great deal of time. Researchers now often have to switch off genes one by one to find out which ones are responsible for a particular trait. "Our method allows them to make a targeted selection beforehand. It helps them develop hypotheses more efficiently and find their way through the vast amount of DNA data that is now available."