Efficient iterative virtual screening with Apache Spark and conformal prediction.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Bioinformatics

Background:

• Structure-based virtual screening is crucial for identifying drug candidates.
• Current methods often involve brute-force docking and scoring of large ligand libraries.
• This approach is computationally intensive and time-consuming.

Purpose of the Study:

• To develop an efficient and accurate virtual screening strategy.
• To reduce the computational burden of docking large ligand libraries.
• To improve the speed and efficacy of identifying potential drug candidates.

Main Methods:

• Iterative docking and modeling using support vector machines (SVM) and conformal prediction.
• Utilizing Apache Spark for parallelized docking and modeling.
• Developing a strategy to predict and exclude 'low-scoring' ligands iteratively.

Main Results:

• Conformal prediction based virtual screening (CPVS) reduced the number of docked molecules by an average of 62.61%.
• Achieved an average accuracy of 94% for the top 30 hits across four different targets.
• Demonstrated a speedup factor of 3.7 in the virtual screening process.

Conclusions:

• CPVS offers a significant improvement over traditional virtual screening methods.
• The method is effective in reducing computational costs while maintaining high accuracy.
• The open-source implementation facilitates its application on various computational resources.