TISCalling: leveraging machine learning to identify translational initiation sites in plants and viruses

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Identifying translational initiation sites (TISs) is crucial for discovering novel proteins and peptides.
• Conventional computational methods struggle with systematic TIS identification, especially non-AUG sites in plants, and evaluating mRNA sequence feature importance.

Purpose of the Study:

• To develop a robust framework, TISCalling, for systematic and global identification and ranking of novel TISs across eukaryotes.
• To evaluate the importance of mRNA sequence features for TIS determination in plants and mammals.
• To identify novel viral TISs with high predictive power.

Main Methods:

• TISCalling combines machine learning (ML) models and statistical analysis for TIS identification.
• The framework generalizes and ranks features common to plants and mammals, while identifying kingdom-specific features.
• It achieves high predictive power for novel viral TISs.

Main Results:

• TISCalling identifies and ranks novel TISs across eukaryotes, including kingdom-specific features like mRNA secondary structures and G-nucleotide content.
• The framework demonstrates high predictive power for identifying novel viral TISs.
• It provides prediction scores for putative TISs in plant transcripts, enabling prioritization for validation.

Conclusions:

• TISCalling offers a sequence-aware and interpretable approach for decoding genome sequences and exploring functional proteins.
• The framework is available as a command-line package and web tools for broad accessibility.
• It advances the systematic identification of TISs, particularly in plants and viruses.