TITER: predicting translation initiation sites by deep learning.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Gene expression regulation hinges on translation initiation, a complex process complicated by alternative translation initiation sites (TISs).
• Accurate TIS identification is crucial for understanding gene regulation, but remains challenging due to non-canonical start codons and complex regulatory mechanisms.
• High-throughput sequencing technologies like GTI-seq and QTI-seq offer new avenues for studying translation initiation and developing computational TIS identification methods.

Purpose of the Study:

• To develop an accurate, genome-wide computational framework for predicting translation initiation sites (TISs).
• To leverage QTI-seq data for enhanced TIS prediction by integrating sequence context and codon composition.
• To provide a robust tool for analyzing translation initiation efficiency and regulatory mechanisms.

Main Methods:

• Developed TITER, a deep learning framework utilizing a hybrid neural network to extract sequence features surrounding TISs.
• Integrated TIS codon composition preferences into the prediction model.
• Utilized QTI-seq data for training and validation of the TIS prediction framework.

Main Results:

• TITER significantly outperforms existing state-of-the-art methods in TIS identification accuracy.
• Identified sequence signatures specific to different TIS codons, including a Kozak-sequence-like motif for AUG.
• TITER prediction scores correlate with translation initiation strength in various biological contexts, such as upstream open reading frames and mutational effects.

Conclusions:

• TITER provides a powerful and accurate deep learning-based approach for genome-wide TIS prediction.
• The framework enhances understanding of sequence determinants and regulatory factors influencing translation initiation.
• TITER is available as open-source software, facilitating further research in gene expression regulation.