DeepGenGrep

DeepGenGrep identifies genomic signals and regions (GSRs) from DNA sequences using a hybrid deep learning architecture to improve annotation of gene structure, regulation, and function.

Key Features:

• Hybrid Neural Network Architecture: Combines a three-layer convolutional neural network (CNN) and a two-layer long short-term memory (LSTM) network to capture local patterns and sequential dependencies and learn feature representations from DNA sequences.
• Multi-GSR Prediction: Systematically identifies multiple genomic signals and regions, including polyadenylation signals, translation initiation sites, and splice sites.
• Cross-Species Benchmarking: Evaluated on Homo sapiens, Mus musculus, Bos taurus, and Drosophila melanogaster and demonstrated superior performance compared to several state-of-the-art approaches.
• High-Throughput Analysis: Facilitates high-throughput, cost-effective prediction and annotation of eukaryotic genomes.

Scientific Applications:

• Gene Regulation and Functional Genomics: Prediction of GSRs to inform studies of gene structure, transcriptional and post-transcriptional regulation.
• Genome Annotation: Annotation of polyadenylation signals, translation initiation sites, and splice sites in eukaryotic genomes including human, mouse, cattle, and Drosophila.
• Comparative Genomics and Method Benchmarking: Cross-species analyses to compare genomic signals and to benchmark prediction performance against existing methods.

Methodology:

DeepGenGrep employs a hybrid deep learning framework combining a three-layer CNN and a two-layer LSTM to capture spatial and temporal patterns in genomic sequences for GSR prediction.