CapsNet-TIS: Predicting translation initiation site based on multi-feature fusion and improved capsule network

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate identification of translation initiation sites (TIS) is essential for understanding gene regulation, transcription, and translation.
• Existing models struggle to fully extract TIS sequence features and capture complex hierarchical relationships.

Purpose of the Study:

• To propose a novel predictor, CapsNet-TIS, for enhanced TIS identification.
• To improve the feature extraction and sequence modeling capabilities for TIS prediction.

Main Methods:

• Employed four encoding methods: One-hot, physical structure property (PSP), nucleotide chemical property (NCP), and nucleotide density (ND).
• Utilized multi-scale convolutional neural networks for feature fusion.
• Implemented an improved capsule network with residual blocks, channel attention, and BiLSTM for classification.

Main Results:

• CapsNet-TIS demonstrated superior performance in TIS prediction across human, mouse, bovine, and fruit fly datasets.
• Ablation experiments confirmed the effectiveness of individual components of the CapsNet-TIS model.
• Comparative analysis showed improved results over existing TIS prediction models.

Conclusions:

• CapsNet-TIS effectively extracts TIS sequence information and models complex hierarchical features.
• The proposed model offers a significant advancement in TIS identification accuracy.
• This approach holds promise for deeper insights into gene expression regulation.