EnsembleSplice

EnsembleSplice predicts donor and acceptor splice sites in genomic sequences using an ensemble of convolutional neural networks to improve splice site detection accuracy.

Key Features:

• Ensemble learning architecture: An ensemble composed of four distinct convolutional neural networks (CNNs) that integrates multiple CNN architectures for robust splice site prediction.
• High prediction accuracy: Achieves 94.16% accuracy for acceptor splice sites and 95.97% accuracy for donor splice sites on Homo sapiens datasets.
• Low error rates: Reports a 4.03% error rate for donor splice sites and a 5.84% error rate for acceptor splice sites on Homo sapiens datasets.
• Cross-validation: Employs five-fold cross-validation to assess consistency and reliability of prediction performance.
• Diverse dataset evaluation: Evaluated on multiple genomic datasets, including two Homo sapiens datasets and an Arabidopsis thaliana dataset.
• Model training and evaluation: Trains and tests ensembles composed of CNNs and deep neural networks (DNNs) and uses evaluation and diversity metrics for model selection.

Scientific Applications:

• Genomics: Supports understanding of gene expression regulation through accurate splice site identification.
• Bioinformatics: Enables precise identification of splice site regions to improve genomic sequence analyses.
• Computational biology: Aids identification of genetic variations and exploration of potential therapeutic targets via improved splice site detection.

Methodology:

Trained and tested ensembles of CNNs and DNNs using five-fold cross-validation and evaluation/diversity metrics to identify optimal model configurations.