iATC-NRAKEL

iATC-NRAKEL predicts first-level Anatomical Therapeutic Chemical (ATC) classes of drugs by extracting network-derived features and applying multi-label machine learning for drug class identification.

Key Features:

• Novel multi-label classification: Uses a multi-label classification strategy tailored to predict first-level ATC classes.
• Enhanced feature extraction: Aggregates drug association data from STITCH and KEGG into seven distinct drug networks for feature generation.
• Network embedding with Mashup: Applies the Mashup network embedding algorithm to convert network information into numerical feature representations.
• Integration with RAKEL and SVMs: Feeds Mashup-derived features into the RAndom k-labELsets (RAKEL) algorithm using support vector machines as base classifiers for multi-label prediction.
• Performance metrics: Achieved 76.56% accuracy and 74.51% absolute true rate in 10-fold cross-validation on a benchmark dataset of 3,883 drugs.
• Network contribution analysis: Performs analysis of each network's contribution to the overall prediction performance.

Scientific Applications:

• Drug repositioning: Facilitates identification of candidate existing drugs for new indications by providing first-level ATC class predictions.
• Drug discovery: Assists in prioritizing compounds and interpreting therapeutic class signals in drug discovery workflows.
• Pharmacological classification: Enables large-scale ATC class annotation of compound libraries for pharmacological and chemoinformatics studies.

Methodology:

Organize drug association data from STITCH and KEGG into seven networks, apply Mashup for network embedding to produce features, input features into RAKEL with support vector machines for multi-label classification, and evaluate using 10-fold cross-validation on a 3,883-drug benchmark with network contribution analysis.