AGRAMP: machine learning models for predicting antimicrobial peptides against phytopathogenic bacteria

Area of Science:

• Bioinformatics and computational biology
• Agricultural science
• Microbiology

Background:

• Antimicrobial peptides (AMPs) show promise as alternatives to traditional antibiotics for controlling plant pathogens.
• Experimental identification of AMPs is time-consuming and resource-intensive.
• Developing efficient methods for AMP discovery is crucial for sustainable agriculture.

Purpose of the Study:

• To develop and validate machine learning models for predicting antimicrobial peptides (AMPs) active against plant pathogenic bacteria.
• To identify potential AMPs from the citrus genome using a bioinformatics approach.
• To provide a tool for accelerating the discovery of novel AMPs for agricultural applications.

Main Methods:

• Utilized N-gram representations of peptide sequences with reduced amino acid alphabets.
• Employed machine learning models trained and evaluated using 5-fold cross-validation.
• Applied models to predict AMPs from citrus intergenic regions and small open reading frames (ORFs).

Main Results:

• Machine learning models achieved prediction accuracies ranging from 0.72 to 0.91.
• Approximately 7% of peptides from intergenic regions and whole genome datasets were predicted as probable AMPs.
• Experimental validation confirmed the antimicrobial activity of selected predicted AMPs against *Spiroplasma citri*.

Conclusions:

• Machine learning models effectively predict AMPs active against plant pathogens, demonstrating significant potential for agricultural use.
• Key features for AMP prediction include hydrophobic and positively charged amino acid residues, and aggregation propensity.
• The developed model, accessible via the AGRAMP server, facilitates the development of AMP-based strategies for plant disease management.