A new structure prediction model has mapped 500 previously unsolved proteins

Elucidate the structures of proteins secreted by plant pathogens with machine learning-based structure prediction tools. Machine learning and plant-pathogen interaction usually have a black box. When predicting primary input sequences to protein structures, we don’t know exactly what’s going on. Likewise, we do not fully understand the complex interplay at the interface of plants and pathogens. The box in the middle captures the complexity of this black box. Credit: Kyungyong Seong and Ksenia V. Krasileva

Scientists at the University of California, Berkeley recently published work that lays the groundwork for new ways of thinking about pathogen evolution. “Our research highlights that model-free modeling that uses machine learning is indeed superior to model-based modeling for the secreted proteins of the destructive fungal pathogen. Magnaporthe oryzae, said Kyungyong Seong, first author of the article published in the MPMI newspaper.

Pathogens use virulence factors known as effectors, which are important for the survival of the pathogen. Homology modeling is one of the most widely used methods, but it requires the use of resolved effector structure models and resolving all effector structures is too daunting a task. There are too many effector proteins encoded in pathogen genomes to simply rely on the experimental resolution of each structure.

Seong and his colleague Ksenia V. Krasileva used a new structure prediction method capable of modeling 500 secreted proteins previously not predicted by the model-based method.

“About 70% of the 1,854 secreted proteins were modeled in our study, and their structures provide an additional layer of effector information based on their similarity to each other or to other resolved protein structures,” said said Krasileva. “We demonstrate that new methods of structure prediction apply well to the problem of deciphering pathogen virulence factors and other secreted proteins that often have little sequence similarity to each other or to other proteins. “

This new method allows scientists to map thousands of secreted proteins and establish the missing evolutionary connection between them. “We believe our research was the first to apply the concept of structural genomics to a plant pathogen in the new era of machine learning structure prediction,” Seong said.

“As the accuracy of structure prediction improves, it will become more common to see papers that incorporate large-scale protein structure prediction data,” Krasileva predicted. “Our paper may spark ideas about how to use this data, leading some scientists to explore the opportunities before others.”

They also discovered that there are many new structurally similar effectors unrelated to the sequence in M. oryzae, and structurally similar effectors are found in other plant pathogens. This suggests that pathogens may rely on a set of effectors that are generally native but have diverged widely in sequence during evolution to infect plants.

Reference: “Computational structural genomics uncovers common folds and novel families in the secretome of fungal plant pathogens Magnaporthe oryzae” by Kyungyong Seong and Ksenia V. Krasileva, November 10, 2021, MPMI newspaper.
DOI: 10.1094/MPMI-03-21-0071-R

Comments are closed.