CNNSplice: Robust models for splice site prediction using convolutional neural networks

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate splice site identification is crucial for gene annotation and understanding biological function in eukaryotes.
• Existing splice site detection tools often lack transferability and efficiency across different organisms.

Purpose of the Study:

• To develop and evaluate CNNSplice, a deep convolutional neural network-based tool for predicting splice sites.
• To assess the performance and generalizability of CNNSplice compared to existing methods.

Main Methods:

• Utilized deep convolutional neural networks (CNNs) for splice site prediction.
• Employed a five-fold cross-validation technique for model selection.
• Tested models on balanced and imbalanced datasets across five different organisms.

Main Results:

• CNNSplice models demonstrated superior performance compared to existing methods on splice site prediction.
• Generality tests confirmed CNNSplice's ability to predict splice sites in new or poorly annotated genomes.
• The tool offers improved prediction accuracy, interpretability, and generalizability.

Conclusions:

• CNNSplice provides a robust and broadly applicable solution for splice site identification.
• The developed web server offers public access to this advanced tool for the research community.