PSOPIA

PSOPIA predicts protein-protein interactions using an Averaged One-Dependence Estimators (AODE) classifier that integrates three sequence-based features to infer interaction likelihoods from homologous protein pairs.

Key Features:

• AODE algorithm: Employs the Averaged One-Dependence Estimators (AODE) algorithm to combine feature evidence for PPI prediction.
• FSeq (Sequence Similarity): Measures sequence similarities between a query protein pair and known interacting protein pairs.
• FDom (Domain Pair Propensities): Assesses statistical propensities of domain pairs observed in interacting proteins.
• FNet (Network Proximity): Calculates the sum of edge weights along the shortest path between homologous proteins within a known PPI network to quantify proximity.
• Feature Space Symmetry: Defines feature vectors in a half-space of a symmetrical high-dimensional feature space to ensure independence from protein order.
• Validation and Performance: Validated by 10-fold cross-validation on a human PPI dataset with randomly sampled negative data, achieving AUC 0.79 and pAUC0.5% 0.16, and outperforming other homology-based methods on independent datasets.

Scientific Applications:

• Identification of novel PPIs: Predicts candidate protein-protein interactions for experimental follow-up.
• Estimation of complete PPI networks: Aids reconstruction and estimation of comprehensive cellular or organismal PPI networks from sequence data.
• Prediction for proteins of unknown structure: Infers interaction likelihoods using sequence-derived features when tertiary structures are unavailable.
• Functional and pathway analysis: Supports exploration of protein functions, pathways, and cellular mechanisms by providing predicted interaction partners.

Methodology:

Integrates three sequence-derived features (FSeq, FDom, FNet) using the AODE algorithm; feature vectors are constrained to a half-space of a symmetrical high-dimensional feature space to ensure order independence; performance was assessed by 10-fold cross-validation on a human PPI dataset with randomly sampled negative data and evaluated using AUC and pAUC0.5%, with comparisons to other homology-based methods on independent datasets.