EditPredict: Prediction of RNA editable sites with convolutional neural network

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• RNA editing is crucial for biological processes, with millions of human events identified and more anticipated.
• Accurate prediction of RNA editing sites is essential for understanding gene function and disease.

Purpose of the Study:

• To develop and validate Convolutional Neural Network (CNN) models for predicting human RNA editing events.
• To create a user-friendly web service, EditPredict, for RNA editing prediction across various genomes.

Main Methods:

• Construction of CNN models to predict RNA editing in Alu and non-Alu regions.
• Validation using a dataset from CRISPR/Cas9 knockout of the ADAR1 enzyme.
• Development of the EditPredict web service integrating reference and personal genomes.

Main Results:

• CNN models achieved high validation accuracies: 99.5% for Alu regions and 93.6% for non-Alu regions.
• EditPredict supports prediction on reference genomes, personal genomes with single nucleotide variants, and multiple model organisms.
• The tool can predict novel RNA editing events and aid in filtering RNA-Seq data.

Conclusions:

• CNN models provide a robust method for predicting human RNA editing events.
• EditPredict is a versatile tool for RNA editing prediction, applicable to diverse genomic contexts and species.
• This work advances the discovery and analysis of RNA editing sites.