STS-NLSP

STS-NLSP predicts membrane transporter substrate specificity using a network-based multi-label classification framework integrating structural and semantic similarity features.

Key Features:

• Network-Based Label Space Partition (NLSP): Applies similarity-based network construction and community detection to partition label space and build cluster-specific ensemble classifiers for 13 transporter substrate classes, including ATP-binding cassette and solute carrier families.
• Hybrid Feature and Correlation Modeling: Integrates 2D structural fingerprints with chemical ontology–based semantic similarity and quantifies label correlations using Cramér's V statistics.

Scientific Applications:

• Transporter-Substrate Specificity Prediction: Supports drug discovery, pharmacokinetics (ADME) analysis, and investigation of drug resistance mechanisms in cancer by predicting interactions between compounds and membrane transporters.

Methodology:

STS-NLSP combines molecular fingerprint descriptors and semantic similarity features to construct a similarity network, partitions correlated labels using community detection algorithms, trains ensemble multi-label classifiers for each label cluster, and evaluates performance using jackknife testing and iterative stratification-based cross-validation.