PluriPred

PluriPred predicts pluripotency-associated proteins from protein sequences using machine learning and sequence alignment to identify core and extended core pluripotent proteins in human and mouse proteomes.

Key Features:

• Machine learning and alignment methods: Employs Support Vector Machine (SVM), Naïve Bayes (NB), Random Forest (RF) classifiers and sequence alignment through BLAST and PSI-BLAST, with the SVM+PSI-BLAST combination identified as the most effective model.
• Performance metrics: The SVM-PSI-BLAST model achieved sensitivity 77.40%, specificity 79.72%, accuracy 79.2%, and area under the ROC curve (AUC) 0.82, validated via 5-fold cross-validation.
• Prediction confidence: Reports confidence scores derived from SVM score distributions of the training dataset and p-values from BLAST analyses.
• Independent validation: Model performance was validated against independent datasets derived from high-throughput studies.

Scientific Applications:

• Discovery of pluripotent proteins: Predicted 233 novel core and 323 novel extended core pluripotent proteins in the mouse proteome and 167 novel core and 385 novel extended core pluripotent proteins in the human proteome.
• Functional network analysis: Identifies proteins involved in protein-protein networks relevant to stem cell biology, cancer research, and developmental biology.

Methodology:

Applies SVM, Naïve Bayes, and Random Forest classifiers combined with BLAST/PSI-BLAST sequence alignment, evaluates models using 5-fold cross-validation, and computes confidence scores from SVM score distributions and BLAST p-values.