ExamPle: explainable deep learning framework for the prediction of plant small secreted peptides

Area of Science:

• Plant Biology
• Bioinformatics
• Computational Biology

Background:

• Plant Small Secreted Peptides (SSPs) are crucial for plant growth, development, and interactions with microbes.
• Identifying SSPs is vital for understanding their functions, but current machine learning methods rely on handcrafted features, limiting predictive performance.
• Existing methods often overlook latent feature representations, hindering accurate SSP prediction.

Purpose of the Study:

• To develop a novel deep learning model, ExamPle, for explainable prediction of plant SSPs.
• To improve the accuracy and efficiency of SSP identification compared to existing machine learning approaches.
• To provide insights into the sequential characteristics and amino acid contributions associated with SSP functions.

Main Methods:

• Developed ExamPle, a deep learning model utilizing Siamese networks and multi-view representation.
• Employed in silico mutagenesis experiments to analyze sequential characteristics and amino acid contributions.
• Benchmarked ExamPle against existing methods for plant SSP prediction.

Main Results:

• ExamPle significantly outperforms existing methods in predicting plant SSPs.
• The model demonstrates strong feature extraction capabilities.
• Identified that the peptide's head region and specific sequential patterns are key to SSP functions.

Conclusions:

• ExamPle offers a powerful and explainable tool for predicting plant SSPs.
• The model facilitates the discovery of functional characteristics within SSPs.
• ExamPle is expected to advance research in plant growth regulation and the design of effective plant SSPs.