Three- and four-class classification models for P-glycoprotein inhibitors using counter-propagation neural networks

Area of Science:

• Pharmacology
• Computational Chemistry
• Biochemistry

Background:

• P-glycoprotein (P-gp) is an ATP binding cassette (ABC) transporter crucial for xenobiotic protection and a key factor in multi-drug resistance (MDR) during cancer chemotherapy.
• Development of P-gp inhibitors is a major strategy to reverse MDR and enhance cytotoxic drug efficacy, but previous generations faced challenges with selectivity, solubility, and pharmacokinetics.

Purpose of the Study:

• To develop accurate classification models for identifying specific P-gp inhibitors.
• To create models capable of distinguishing between P-gp inhibitors, CYP 3A inhibitors, and co-inhibitors.

Main Methods:

• Generation of classification models using counter-propagation neural networks (CPG-NN).
• Development of three models (SION, SIO, SIN) for classifying 'true' P-gp inhibitors.
• Development of three models (CPBN, CPB1, CPN) for multi-class classification involving P-gp and CYP 3A inhibitors.

Main Results:

• Models demonstrated high accuracy in classifying compounds on both test and external datasets.
• The classification models provided insights into compound bioactivities against multiple targets, including P-gp and CYP 3A.
• Successfully distinguished between P-gp inhibitors, CYP 3A inhibitors, and compounds inhibiting both.

Conclusions:

• Counter-propagation neural networks are effective for developing accurate multi-target classification models for drug discovery.
• These models offer a valuable tool for understanding compound interactions with P-gp and CYP 3A, aiding in the development of more effective cancer therapies.
• The study highlights the potential for improved drug design by considering multiple target interactions.