Using a Graph Convolutional Neural Network Model to Identify Bile Salt Export Pump Inhibitors.

Area of Science:

• Pharmacology
• Computational Chemistry
• Toxicology

Background:

• The bile salt export pump (BSEP) is crucial for eliminating bile salts from liver cells.
• BSEP inhibition can cause bile salt accumulation, leading to cholestasis and drug-induced liver injury.
• Identifying BSEP inhibitors is vital for assessing chemical safety and understanding drug liabilities.

Purpose of the Study:

• To develop predictive machine learning models for identifying potential BSEP inhibitors using public data.
• To evaluate the efficacy of a graph convolutional neural network (GCNN) approach combined with multitask learning for BSEP inhibition prediction.
• To compare the performance of GCNN-based single-task and multitask models, especially in scenarios with limited bioactivity data.

Main Methods:

• Utilized publicly available data to train and validate machine learning models.
• Employed a graph convolutional neural network (GCNN) architecture for predictive modeling.
• Implemented multitask learning to enhance model performance and address data limitations.

Main Results:

• The developed GCNN model achieved a cross-validation receiver operating characteristic area under the curve of 0.86, outperforming other machine learning approaches.
• Multitask GCNN models demonstrated superior performance compared to single-task models.
• Multitask learning effectively addressed challenges associated with limited data availability in bioactivity modeling.

Conclusions:

• The multitask GCNN-based BSEP model is a valuable tool for prioritizing potential BSEP inhibitors in early drug discovery.
• This computational approach aids in the risk assessment of chemicals by predicting their potential to inhibit BSEP.
• The study highlights the utility of multitask learning in developing robust predictive models for targets with scarce data.