DPProm: A Two-Layer Predictor for Identifying Promoters and Their Types on Phage Genome Using Deep Learning

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Phage genome annotation requires robust bioinformatics tools due to the increasing number of sequenced phage genomes.
• Promoters are crucial DNA regulatory regions for gene transcription, making their accurate identification essential.
• Existing computational methods for promoter identification need enhancement for phage-specific applications.

Purpose of the Study:

• To develop a novel, accurate, and robust computational model for identifying phage promoters and classifying their types (host or phage).
• To integrate sequence and handcrafted features using a deep learning approach for improved promoter prediction.
• To create a user-friendly web tool for practical application in phage genome analysis.

Main Methods:

• Introduction of DPProm, a two-layer deep neural network model (DPProm-1L and DPProm-2L).
• DPProm-1L utilizes a dual-channel CNN ensemble for sequence and handcrafted features (free energy, GC content, cumulative skew, Z curve) to distinguish promoters from non-promoters.
• DPProm-2L, also CNN-based, predicts promoter types. A sequence processing workflow with sliding windows and merging modules enables whole-genome prediction.

Main Results:

• DPProm demonstrates superior performance compared to state-of-the-art methods in phage promoter identification.
• The model significantly reduces the false positive rate in whole-genome promoter predictions.
• Experimental validation confirms the effectiveness of the integrated feature approach and the two-layer architecture.

Conclusions:

• DPProm offers a powerful and effective solution for identifying phage promoters and their types, addressing a critical need in phage bioinformatics.
• The developed web tool provides accessible functionality for researchers, facilitating phage genome annotation.
• This work advances computational methods for regulatory element identification in microbial genomes.