Topology-Based and Conformation-Based Decoys Database: An Unbiased Online Database for Training and Benchmarking...

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Machine-learning-based scoring functions (MLSFs) show promise for enhancing binding affinity prediction and structure-based virtual screening (SBVS).
• Accurate MLSF development necessitates large, diverse, and unbiased datasets, which are currently lacking.
• Existing datasets often contain hidden biases and insufficient data, hindering MLSF performance.

Purpose of the Study:

• To develop a novel, large-scale, and unbiased decoy database for training and evaluating MLSFs in SBVS.
• To address the limitations of existing datasets in terms of size, diversity, and bias.
• To provide a valuable resource for the computational chemistry and drug discovery communities.

Main Methods:

• Collected biological targets and active ligands from scientific literature and established databases.
• Generated and debiased decoys using conditional recurrent neural networks and molecular docking techniques.
• Compiled the Topology-based and Conformation-based Decoys Database (ToCoDDB).

Main Results:

• ToCoDDB is the largest unbiased decoy database to date, containing 2.4 million decoys across 155 targets.
• The database provides detailed information and performance benchmarks for each target, facilitating MLSF development.
• An online decoy generation function was implemented, expanding the database's utility.

Conclusions:

• ToCoDDB offers a significant advancement for the development and validation of accurate MLSFs for SBVS.
• The database and its online generation tool will accelerate drug discovery efforts.
• ToCoDDB is freely accessible, promoting wider research and application in computational drug design.