Spliceator: multi-species splice site prediction using convolutional neural networks.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate splice site prediction is crucial for eukaryotic genome annotation.
• Current Deep Learning (DL) methods often rely on model organism data, limiting their application to non-model organisms.
• There is a need for robust DL-based splice site predictors applicable across a wider range of species.

Purpose of the Study:

• To develop a novel Deep Learning method for ab initio splice site prediction.
• To create a tool that performs accurately across diverse species, including non-model organisms.
• To improve the efficiency and accuracy of genome annotation.

Main Methods:

• Development of Spliceator, a convolutional neural network (CNN) based predictor.
• Training the CNN on a carefully curated dataset comprising validated splice site data from over 100 organisms.
• Benchmarking Spliceator's performance against existing methods on independent datasets.

Main Results:

• Spliceator demonstrates high prediction accuracy, ranging from 89% to 92%, on independent benchmarks.
• The method shows consistent high performance across various species, including human, fish, fly, worm, plant, and protist.
• Spliceator outperforms existing methods in splice site prediction accuracy.

Conclusions:

• Spliceator is an effective Deep Learning method for predicting splice sites in a broad spectrum of organisms.
• The tool provides accurate splice site predictions for both model and non-model organisms.
• Spliceator offers a valuable advancement for eukaryotic genome annotation, particularly for understudied species.