mamp-ml: A deep learning approach to epitope immunogenicity in plants

Area of Science:

• Plant biology
• Immunology
• Bioinformatics

Background:

• Eukaryotes use surface receptors to detect biomolecules, including pathogens, to trigger immunity.
• Vast sequence variation in plant receptors and pathogen ligands creates an experimental bottleneck for studying immune interactions.
• Understanding these interactions is crucial for enhancing plant defense mechanisms.

Purpose of the Study:

• To develop a machine learning framework, mamp-ml, for predicting plant receptor-ligand interactions.
• To computationally screen large numbers of receptor-ligand combinations.
• To provide a tool for engineering plant immune systems.

Main Methods:

• Leveraged two decades of functional data and the ESM-2 protein language model.
• Developed a machine learning pipeline integrating receptor-ligand features.
• Trained a model to predict immunogenic outcomes.

Main Results:

• Achieved 73% prediction accuracy on a held-out test set.
• Demonstrated predictive power even without experimental protein structures.
• Enabled high-throughput screening of leucine-rich repeat (LRR) receptor-ligand combinations.

Conclusions:

• mamp-ml offers a robust computational framework for predicting plant immune responses.
• The approach facilitates rapid identification of functional receptor-ligand pairs.
• Provides a foundation for the rational design and engineering of plant immunity.