PETs

PETs predicts protein-protein interaction (PPI) sites at the amino-acid level using an extremely-randomized trees machine learning framework combined with a cluster-based sampling strategy to improve prediction accuracy and stability.

Key Features:

• Amino-acid-level PPI site prediction: Predicts residue-level protein-protein interaction sites for proteins.
• Extremely-Randomized Trees Algorithm: Employs extremely-randomized trees for classification of PPI site residues.
• Cluster-Based Sampling Strategy: Divides the training dataset into subsets using predefined features, clusters those subsets with K-means, and selects samples from each cluster to ensure diverse representation of significant features.
• Model Stability Enhancement: Uses cluster-based sampling to reduce variability from random sampling and improve consistency across training iterations.

Scientific Applications:

• Protein-Protein Interaction Prediction: Identification of interaction sites between proteins to support studies of cellular processes and disease mechanisms.
• Model Stability and Accuracy Assessment: Generation of more consistent and accurate residue-level PPI predictions across different training runs.

Methodology:

The method segments the dataset into subsets using predefined features, clusters subsets via K-means, selects samples from each cluster to ensure diverse representation, and trains an extremely-randomized trees classifier on the sampled data.