NpPred

NpPred predicts nuclear proteins from amino acid sequences to identify nuclear localization and nuclear versus non-nuclear domains relevant to chromosomal maintenance, gene expression, and RNA processing.

Key Features:

• Data Foundation: Built on a non-redundant dataset of 2710 nuclear and 7662 non-nuclear proteins for training and validation.
• SVM Modules: Employs support vector machines using amino acid composition, dipeptide composition, and split amino acid composition (SAAC).
• Split Amino Acid Composition (SAAC): Computes amino acid composition of the N-terminus separately from the remainder of the protein and achieved an MCC of 0.66.
• HMM Profile: Implements a hidden Markov model-based module/profile to identify exclusively nuclear and non-nuclear domains within proteins.
• Hybrid Module: Integrates the SVM module with the HMM profile, achieving an MCC of 0.87 and an accuracy of 94.61%.
• Validation: Evaluated using five-fold cross-validation and tested on blind/independent datasets where it outperformed existing methods.
• Proteome Estimates: Estimates nuclear protein proportions for Saccharomyces cerevisiae (31.51%), Caenorhabditis elegans (21.89%), Drosophila melanogaster (26.31%), mouse (25.72%), and human (24.95%).

Scientific Applications:

• Nuclear Protein Prediction: Sequence-based discrimination of nuclear versus non-nuclear proteins for genomics and proteomics studies.
• Domain Identification: Detection of nuclear and non-nuclear protein domains to inform studies of subcellular localization and function.
• Proteome Analysis: Estimation of nuclear protein prevalence across proteomes for comparative genomics and proteomics analyses.

Methodology:

Uses SVMs with amino acid composition, dipeptide composition and SAAC (N-terminus separate), an HMM-based domain profile, a hybrid SVM+HMM module, and was evaluated by five-fold cross-validation and blind/independent dataset testing.