TPpred 1.0

TPpred 1.0 predicts organelle targeting peptides and their cleavage sites to determine the localization and sequence maturation of nuclear-encoded proteins imported into mitochondria and plastids.

Key Features:

• Algorithm: Implements Grammatical-Restrained Hidden Conditional Random Fields (GR-HCRFs) to model sequence patterns of targeting peptides.
• Training data: Trained on a non-redundant dataset of 297 experimentally validated proteins containing targeting peptides.
• Peptide prediction: Identifies targeting peptides for mitochondrial and plastid import from nuclear-encoded protein sequences.
• Cleavage site prediction: Predicts cleavage sites with average positional errors of 7 residues for mitochondrial proteins and 15 residues for plastidic proteins.
• Performance metrics: Reported accuracy of 96%, Matthews correlation index (MCI) of 0.58, and a false-positive rate of 3.0%.
• Proteome-level estimates: Applied to human, Arabidopsis thaliana, and yeast proteomes, estimating ~4–9% of sequences contain targeting peptides.
• Annotation alignment: Predictions align with experimental subcellular localization annotations and inform sequence maturation assessment.

Scientific Applications:

• Proteome annotation: Assignment of mitochondrial and plastid targeting signals during genome-scale protein annotation.
• Subcellular localization studies: Prediction-based inference of protein localization for experimental design and validation.
• Large-scale proteomics: Low false-positive rate makes the method suitable for proteome-wide surveys of targeting peptides.
• Comparative genomics: Estimation of targeting peptide prevalence across species such as human, Arabidopsis thaliana, and yeast.

Methodology:

Uses Grammatical-Restrained Hidden Conditional Random Fields (GR-HCRFs) trained on a non-redundant set of 297 experimentally validated proteins containing targeting peptides.