Virtual screening system for finding structurally diverse hits by active learning.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Virtual screening is crucial for identifying potential drug candidates.
• Traditional similarity-based methods can be limited in finding structurally diverse compounds.
• Active learning strategies offer a promising approach to enhance virtual screening efficiency.

Purpose of the Study:

• To develop and evaluate novel virtual screening strategies based on active learning.
• To compare the performance of "query by bagging" (QBag) and "query by bagging with descriptor-sampling" (QBagDS) against conventional methods.
• To assess the impact of descriptor sampling and prior knowledge on hit identification.

Main Methods:

• Devised two active learning-based virtual screening strategies: QBag and QBagDS.
• QBag generates multiple structure-activity relationship rules via bagging and iteratively selects compounds.
• QBagDS incorporates descriptor sampling to enhance structural diversity of identified hits.

Main Results:

• Simulation experiments and clustering analysis demonstrated that QBag and QBagDS outperform similarity-based screening.
• The combination of descriptor sampling and prior knowledge significantly improved the identification of numerous hits.
• Application of QBagDS to novel hit finding resulted in 4 out of 10 selected compounds exhibiting high inhibition.

Conclusions:

• QBag and QBagDS are effective virtual screening strategies for identifying diverse and potent drug candidates.
• Descriptor sampling and prior knowledge integration are key factors for maximizing hit discovery.
• These advanced methods show significant potential for accelerating novel drug discovery pipelines.