DeepLRR: An Online Webserver for Leucine-Rich-Repeat Containing Protein Characterization Based on Deep Learning

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Leucine-rich repeat (LRR) proteins are crucial in biological processes.
• Predicting LRR unit number and location is challenging due to sequence variability.
• Current prediction methods require improvement, particularly similarity-based approaches.

Purpose of the Study:

• To develop an accurate method for predicting LRR units in proteins.
• To improve the identification of plant disease resistance proteins.
• To re-annotate and analyze LRR-RLK genes in key plant genomes.

Main Methods:

• Developed DeepLRR, a convolutional neural network (CNN) model.
• Utilized specific LRR features for prediction.
• Compared DeepLRR against six existing methods on a dataset of 572 LRR proteins.

Main Results:

• DeepLRR achieved superior performance with a high overall F1 score.
• Successfully identified plant disease-resistance proteins (NLR, LRR-RLK, LRR-RLP) and non-canonical domains.
• Re-annotated 223, 191, and 183 LRR-RLK genes in Arabidopsis, rice, and tomato, respectively.
• Discovered gene cluster structures for a significant percentage of LRR-RLK genes in these plants.

Conclusions:

• DeepLRR offers a significant advancement in predicting LRR units.
• The method facilitates the identification and analysis of plant disease resistance genes.
• Provides insights into the evolutionary relationships and genomic organization of LRR-RLK genes, aiding receptor discovery.