Practical Model Selection for Prospective Virtual Screening | JoVE Visualize

Area of Science:

Computational chemistry
Drug discovery
Bioinformatics

Background:

Virtual screening (VS) aids in identifying potential drug candidates by prioritizing compounds for experimental testing.
The effectiveness of VS algorithms varies significantly based on the dataset and evaluation methods used.
Protein-protein interactions (PPIs) like PriA-SSB and RMI-FANCM are challenging targets in drug discovery.

Purpose of the Study:

To develop a strategy for selecting the optimal virtual screening algorithm for prospective compound prioritization.
To evaluate a diverse set of ligand-based machine learning and docking-based VS approaches.
To compare the performance of different VS methods on specific protein targets relevant to drug discovery.

Main Methods:

Investigated various ligand-based machine learning algorithms, including random forest and neural networks.
Employed docking-based approaches for virtual screening.
Assessed algorithm performance on two specific protein targets: PriA-SSB and RMI-FANCM.
Validated findings using a library of 22,434 new molecules against the PriA-SSB target.

Main Results:

A random forest algorithm was identified as the most effective VS approach for the studied targets.
The selected random forest model successfully recovered 37 out of 54 active compounds from the test library for PriA-SSB.
Sophisticated models like multi-task neural networks did not consistently outperform simpler methods in prospective screening.
Performance on public datasets or synthetic benchmarks does not always predict success in specific experimental assays.

Conclusions:

The optimal virtual screening algorithm choice is context-dependent and requires careful evaluation for specific biological targets.
Random forest models offer a robust and effective strategy for prospective compound prioritization in drug discovery.
The study highlights the limitations of relying solely on benchmark performance for selecting VS tools.
A tailored approach to VS algorithm selection is crucial for maximizing the efficiency of experimental screening campaigns.