maxent-ppi

maxent-ppi evaluates protein-protein interactions using maximum entropy and support vector machine classifiers to predict co-complex associations from Gene Ontology (GO) and InterPro annotations.

Key Features:

• Ontology-Based Evaluation: Uses Gene Ontology (GO) to standardize and record experimental findings related to protein-protein interactions.
• Dependent Heterogeneous Annotations: Integrates dependent heterogeneous protein annotations drawn from the entire ontology for comprehensive analysis.
• Combinatorial Annotation Construction: Builds PPI annotations combinatorially using corresponding GO terms and InterPro annotations.
• Training Set Design: Employs a high-confidence complex dataset from Saccharomyces cerevisiae as a positive training set and trains classifiers on various training sets.
• Machine Learning Algorithms: Implements Maximum Entropy and Support Vector Machine (SVM) classifiers, each paired with a composite counterpart algorithm.
• Performance Metrics: Reports area under the Receiver Operating Characteristic (ROC) curve ≤0.97 and outperforms the GO-based predictor go2ppi.

Scientific Applications:

• Co-complex Interaction Prediction: Predicts putative co-complex protein interactions from ontology-derived annotations.
• Protein Function and Process Inference: Supports inference of protein function and biological process associations from integrated GO and InterPro annotations.
• Benchmarking of GO-based PPI Methods: Provides comparative performance assessment against GO-based predictors such as go2ppi.

Methodology:

Constructs PPI annotations combinatorially from GO terms and InterPro annotations, and trains Maximum Entropy and SVM classifiers (with composite counterparts) on a high-confidence Saccharomyces cerevisiae complex dataset and additional training sets.