DeepGenGrep: a general deep learning-based predictor for multiple genomic signals and regions.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate annotation of genomic signals and regions (GSRs) is crucial for understanding gene structure, regulation, and function.
• Existing machine learning predictors for in silico identification of GSRs often yield unsatisfactory performance.
• There is a need for more accurate computational methods for GSR prediction.

Purpose of the Study:

• To propose DeepGenGrep, a general deep learning framework for systematic identification of multiple GSRs from genomic DNA sequences.
• To develop a more accurate and effective computational tool for GSR prediction.

Main Methods:

• Developed DeepGenGrep, a deep learning framework utilizing hybrid neural networks.
• Employed a three-layer convolutional neural network (CNN) and a two-layer long short-term memory (LSTM) network.
• Learned effective feature representations from DNA sequences.

Main Results:

• DeepGenGrep demonstrated superior performance compared to state-of-the-art approaches.
• The framework accurately identified polyadenylation signals, translation initiation sites, and splice sites.
• Performance was validated across four eukaryotic species: Homo sapiens, Mus musculus, Bos taurus, and Drosophila melanogaster.

Conclusions:

• DeepGenGrep offers a useful tool for high-throughput and cost-effective identification of potential GSRs in eukaryotic genomes.
• The developed framework advances the accuracy of in silico GSR prediction.
• DeepGenGrep provides a valuable resource for genomic research.