TPpred 3.0

TPpred 3.0 predicts and localizes N-terminal organelle-targeting peptides and their cleavage sites in eukaryotic proteins to annotate mitochondrial and chloroplast targeting and mitochondrial subcompartment localization.

Key Features:

• Detection of Targeting Peptides: Identifies N-terminal targeting peptides that direct nuclear-encoded proteins to mitochondria or chloroplasts using models trained on a non-redundant dataset of experimentally validated sequences.
• Classification as Mitochondrial or Chloroplastic: Distinguishes mitochondrial versus chloroplast-targeting peptides using an N-to-1 Extreme Learning Machine.
• Cleavage-Site Localization: Localizes peptide cleavage sites with organelle-specific Support Vector Machines and an extensive motif-discovery analysis of mitochondrial and chloroplastic proteins.
• Improved Performance: Achieves superior accuracy in detection, classification, and cleavage-site localization compared with state-of-the-art methods and reports a low false-positive rate (3.0%).
• Subcellular Localization Prediction (DeepMito): Predicts mitochondrial subcompartment destinations, including inner membrane, outer membrane, intermembrane space, and matrix, via the DeepMito method.

Scientific Applications:

• Proteome annotation: Annotates proteomes with targeting peptide presence and organelle assignment, particularly for photosynthetic eukaryotes.
• Mature sequence inference: Infers mature protein sequences by predicting targeting-peptide cleavage sites.
• Functional localization: Supports interpretation of protein roles by predicting localization within specific organelle compartments and mitochondrial subcompartments.

Methodology:

Uses Grammatical-Restrained Hidden Conditional Random Fields (used in earlier versions), an N-to-1 Extreme Learning Machine for organelle classification, organelle-specific Support Vector Machines for cleavage-site localization, motif-discovery analysis of mitochondrial and chloroplastic proteins, and DeepMito for mitochondrial subcompartment prediction; earlier GR-HCRF performance reported a Matthews correlation coefficient of 0.58.