ASFP (Artificial Intelligence based Scoring Function Platform): a web server for the development of customized...

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Virtual screening (VS) using molecular docking is crucial for cost-effective drug discovery.
• Current scoring functions (SFs) in docking often lack sufficient accuracy, posing a challenge for reliable predictions.
• Improving the accuracy of SFs is essential for advancing structure-based VS.

Purpose of the Study:

• To introduce ASFP, an integrated web server for creating customized SFs for structure-based VS.
• To enhance the prediction accuracy of SFs through AI and machine learning techniques.
• To provide a powerful tool for developing target-specific and generic SFs for drug discovery.

Main Methods:

• ASFP integrates three modules: descriptor generation (3437 descriptors), AI-based SF construction using machine learning (ML), and an online prediction module.
• The platform utilizes ML techniques to build target-specific SFs based on generated descriptors.
• It offers pre-constructed target-specific SFs and a generic SF for binding affinity prediction.

Main Results:

• ASFP demonstrated high performance in validation studies on benchmark datasets.
• Target-specific SFs achieved an average ROC AUC of 0.973 across 32 targets.
• The generic SF obtained a Pearson correlation coefficient of 0.81 on the PDBbind 2016 core set.

Conclusions:

• The ASFP server provides a robust solution for developing customized SFs in structure-based VS.
• Its AI-driven approach significantly improves the accuracy of molecular interaction predictions.
• ASFP is a valuable tool for accelerating drug discovery through enhanced virtual screening capabilities.