NetStart

NetStart predicts canonical translation initiation sites (TIS) in mRNA transcripts across diverse eukaryotic species, including vertebrates, invertebrates, plants, fungi, and protists, to delineate transitions from non-coding to coding regions for protein synthesis.

Key Features:

• Translation initiation site prediction: Predicts canonical TIS in mRNA transcripts across a wide array of eukaryotic taxa.
• ESM-2 integration: Integrates the ESM-2 protein language model to assess sequence "protein-ness" and inform coding versus non-coding transitions.
• Local start codon context: Incorporates local start codon sequence context around candidate initiation sites in predictions.
• Taxonomical information: Utilizes taxonomical information to adapt predictions across vertebrates, invertebrates, plants, fungi, and protists.
• Training dataset: Trained on sequences from 60 phylogenetically diverse eukaryotic species.
• Transcript–peptide integration: Bridges transcript- and peptide-level information to link mRNA features with peptide-relevant properties.
• Performance: Combines protein language model outputs with local sequence context to achieve state-of-the-art performance in TIS prediction.

Scientific Applications:

• TIS identification: Enables identification of translation initiation sites to study protein synthesis and translational regulation across eukaryotes.
• Integrated transcript–peptide analyses: Facilitates analyses that integrate transcript-level sequence information with peptide-level properties for complex biological prediction tasks.

Methodology:

Integrates the ESM-2 protein language model with local start codon sequence context and taxonomical information; trained on a dataset comprising sequences from 60 phylogenetically diverse eukaryotic species, with ESM-2 used to assess transitions between non-coding and coding regions by evaluating "protein-ness".