on the July 7, 2023
INRAe's GAME team, with support from the MSI, has developed the NERD (Nematode EffectoR Discovery) tool. Through the combined use of various AI techniques, NERD allows for the identification of proteins involved in the parasitic process of crop pests based solely on their sequences.
Producing sufficient agricultural resources to feed a world population estimated at over 9 billion by 2050, while reducing the impact on our planet, is one of the main challenges humanity faces.
Various phytoparasites such as Plant Parasitic Nematodes (PPNs) (microscopic roundworms) and certain species of Oomycetes (filamentous fungus-like organisms) intensify this global food insecurity by causing considerable agricultural losses. In order to infect plants, these organisms rely on an arsenal of proteins called effectors which enable them to manipulate the development, immune response, and physiology of the host plant.
Effectively selecting new candidate effectors in order to study their role in the infectious process is therefore necessary. However, given the diversity of effectors within these pathogens, no simple criterion allows for the efficient discrimination of effector from the rest of the proteins.
To address this challenge, engineers from the MSI, in collaboration with the GAME team (INRAe), have developed NERD (Nematode EffectoR Discovery): a tool that can predict whether a given protein is a potential effector based solely on its sequence. NERD is based on the combined use of supervised and unsupervised approaches distributed in two blocks. The first block relies on the use of an embedding model (Protrans) pre-trained on over 45 million proteins (Uniref50). This 'transformer' is responsible for representing protein sequences as numerical vectors. This representation allows for the extraction of essential protein features that will be used to discriminate effectors from the rest of the proteins in the second block.
This second block, a dense neural network trained on a set of proteins for which it is known whether they are effectors or not, uses this data representation to predict the probability of each analyzed protein being an effector. NERD thus allows for the isolation of a list of candidate effectors from a set of proteins, which can then be validated though experimental biology techniques.
Ultimately, studying the role of these effectors in the parasitic process will lead to development of more effective and environmentally-friendly methods of combating these pathogens.
Partners
MSI – Center of Modeling, Simulation and Interactions
- Dr. Djampa Kozlowski
- Dr. Silvia Bottini
INRAe - Institut National de Recherche pour l'Agriculture, l'Alimentation et l'Environnement (ISA - Institut Sophia Agrobiotech, Game team)
- Dr. Etienne Danchin
