Deep Splicer: A CNN Model for Splice Site Prediction in Genetic Sequences

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• DNA contains genes, which are sequences for protein synthesis, including coding exons and non-coding introns.
• Splice sites are crucial for removing introns during RNA transcription, a key step in gene expression.
• Current biochemical methods for splice site detection are time-consuming and expensive, necessitating computational approaches.

Purpose of the Study:

• To develop an advanced deep learning model, Deep Splicer, for accurate splice site identification in genomic sequences.
• To improve upon the accuracy and reduce false-positive rates of existing computational splice site prediction tools.

Main Methods:

• Development of Deep Splicer, a deep learning model utilizing genomic sequence data.
• Evaluation of Deep Splicer's performance against existing tools like Splice2Deep and Splice Finder.
• Comparative analysis of accuracy and false-positive rates across various species, including humans.

Main Results:

• Deep Splicer demonstrated high accuracy, ranging from 93.55% to 99.66% across different organisms.
• The model achieved significantly lower false-positive rates (0.11% for human, 0.25% for other species) compared to Splice Finder (1-10%).
• Deep Splicer outperformed Splice2Deep in accuracy for human and fruit fly sequences but was slightly surpassed in worm and thale cress.

Conclusions:

• Deep Splicer represents a significant advancement in computational splice site detection.
• The model's high accuracy and low false-positive rate make it a valuable tool for genome annotation.
• Deep Splicer has the potential to accelerate genomic research by improving the efficiency of gene structure analysis.