clinod

clinod predicts nucleolar localization sequences (NoLSs) from protein sequences to identify targeting motifs that direct proteins to the nucleolus.

Key Features:

• Input Format: Accepts sequence lists in FASTA format.
• Prediction Model: Employs an artificial neural network trained on 46 human NoLSs collected from the literature.
• Sequence-level Features: Statistical analysis of training NoLSs shows ~48% basic residues, ~99% predicted solvent accessibility, ~42% predicted α-helix and ~57% predicted coil.
• Output: Produces NoLS predictions and scoring for input sequences.
• Performance: Achieves a true positive rate of 54% with an overall false positive rate of 1.52% (cytoplasmic sequences 0.26%, nucleoplasmic sequences 0.80%, nuclear localization signals 12%).
• Experimental Validation: Ten of the highest-scoring previously unknown NoLSs were confirmed experimentally.

Scientific Applications:

• Proteome-wide Discovery: Identification of novel NoLSs within large datasets such as the complete human proteome.
• Candidate Prioritization: Prioritization of sequence regions for experimental validation of nucleolar targeting.
• Motif Differentiation: Differentiation of NoLSs from other nuclear targeting motifs, including nuclear localization signals.

Methodology:

An artificial neural network trained on 46 human NoLSs was used, combined with statistical analyses of residue composition, predicted solvent accessibility and secondary structure, and evaluated using true positive and false positive rates against cytoplasmic, nucleoplasmic and nuclear localization signal negative sets.