GTB-PPI

GTB-PPI predicts protein-protein interactions from sequence-derived descriptors using gradient tree boosting and L1-regularized logistic regression to support analysis of genetic mechanisms, disease pathogenesis, and drug design.

Key Features:

• Methodology: Integrates gradient tree boosting (GTB) with L1-regularized logistic regression for feature selection.
• Feature Extraction: Constructs initial feature vectors by fusing pseudo amino acid composition (PseAAC), pseudo-position-specific scoring matrix (PsePSSM), reduced sequence and index-vectors (RSIV), and autocorrelation descriptor (AD).
• Feature Selection: Applies L1-regularized logistic regression to select an optimal subset of informative features.
• Model Construction: Builds the predictive model using gradient tree boosting as an ensemble learning technique.
• Datasets: Evaluated on Saccharomyces cerevisiae and Helicobacter pylori datasets and tested on independent sets including Caenorhabditis elegans, Escherichia coli, Homo sapiens, and Mus musculus, and includes one-core and crossover PPI network data such as the CD9 network.

Scientific Applications:

• PPI Prediction Accuracy: Reported validation accuracies of 95.15% for Saccharomyces cerevisiae and 90.47% for Helicobacter pylori.
• Cross-Species Applicability: Applicable to independent test sets from Caenorhabditis elegans, Escherichia coli, Homo sapiens, and Mus musculus.
• Network Analysis: Enables analysis of complex PPI networks including one-core networks for CD9 and crossover PPI networks to investigate interaction patterns.

Methodology:

Initial feature vectors are generated by fusing PseAAC, PsePSSM, RSIV, and AD descriptors; L1-regularized logistic regression performs feature selection; the final classifier is trained using gradient tree boosting.