Profiling prediction of nuclear receptor modulators with multi-task deep learning methods: toward the virtual...

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular biology

Background:

• Nuclear receptors (NRs) are crucial drug targets, but current computational methods lack cross-target knowledge transfer.
• Developing effective NR modulators requires high affinity and specificity, posing a significant challenge.

Purpose of the Study:

• To introduce NR-Profiler, a novel computational framework for predicting potential NR modulators.
• To enhance drug discovery by enabling efficient virtual screening and specificity assessment for NRs.

Main Methods:

• Construction of a comprehensive NR dataset with 42,684 interactions between 42 NRs and 31,033 compounds.
• Development of multi-task deep neural network and graph convolutional neural network models.
• Creation of a consensus model incorporating a selectivity score for specificity measurement.

Main Results:

• The consensus model achieved an AUC of 0.883, outperforming conventional methods in external validation.
• NR-Profiler demonstrated practical utility in virtual screening for NRs.
• A selectivity score was developed to quantify the specificity of NR modulators.

Conclusions:

• NR-Profiler offers a robust and accurate tool for NR-profiling prediction.
• The framework facilitates NR-based drug discovery by improving modulator prediction and specificity assessment.
• A freely available standalone software is provided for broader accessibility and application.