Descriptor Data Bank (DDB): A Cloud Platform for Multiperspective Modeling of Protein-Ligand Interactions

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Protein-ligand (PL) interactions are crucial for biological processes, involving forces like hydrogen bonding and hydrophobic effects.
• Existing models for these interactions are fragmented, limiting accurate prediction of binding affinity.
• A unified approach is needed to leverage diverse hypotheses for improved PL interaction modeling.

Purpose of the Study:

• To introduce the Descriptor Data Bank (DDB), a cloud-based platform for multiperspective modeling of PL interactions.
• To integrate diverse descriptor extraction tools and machine learning for enhanced prediction of binding affinity.
• To develop and validate novel protein-specific descriptors for improved interaction modeling.

Main Methods:

• Developed DDB, an open-access platform hosting descriptor extraction tools and data for PL interaction hypotheses.
• Implemented a machine learning (ML) toolbox for descriptor filtering, analysis, and scoring function (SF) fitting.
• Generated over 2700 descriptors from 16 tools, including novel protein-specific descriptors based on structural alignments.
• Fit SFs using DDB's ML library and evaluated them on drug screening-relevant datasets.

Main Results:

• Multiperspective SFs built with diverse DDB descriptors outperformed single-perspective models by over 15% on average.
• The descriptor filtering module successfully reduced irrelevant and noisy features.
• Novel protein-specific descriptors demonstrably improved the accuracy of SFs.
• DDB facilitates collaborative development and sharing of PL interaction modeling tools and data.

Conclusions:

• Multiperspective modeling using the DDB platform significantly enhances the accuracy of protein-ligand binding affinity prediction.
• The DDB platform provides a valuable resource for the computational biology and drug discovery communities.
• Protein-specific descriptors represent a promising avenue for improving the precision of molecular interaction modeling.