Ligand Based Virtual Screening Using Self-organizing Maps.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Bioinformatics

Background:

• Traditional drug discovery is time-consuming and expensive, relying on extensive in-vitro screening of small molecules against target molecules.
• The vast chemical space for potential ligands makes identifying effective drug candidates a significant challenge.
• Virtual screening computational methods can accelerate drug discovery by reducing the search space and identifying promising lead molecules.

Purpose of the Study:

• To propose a ligand-based virtual screening method utilizing a self-organizing map (SOM), an artificial neural network.
• To enhance the accuracy and efficiency of identifying active drug molecules.
• To develop a graphics processing unit (GPU)-accelerated SOM model for faster computation.

Main Methods:

• Implementation of a ligand-based virtual screening approach using two SOMs to independently predict molecule activity (active/inactive).
• Development of a GPU-based SOM model to leverage computational parallelism for accelerated training and screening.
• Comparison of the SOM-based method against support vector machine and random forest models.

Main Results:

• The proposed SOM technique can classify molecules as active, inactive, or undefined, offering a nuanced prediction.
• The SOM-based approach demonstrated a reduction in false positives and an improvement in recall compared to existing models.
• The GPU-accelerated SOM model significantly improved execution time, enabling the evaluation of large molecule datasets.

Conclusions:

• The SOM-based virtual screening method offers a more accurate and efficient alternative to conventional drug discovery techniques.
• The GPU implementation of SOM provides substantial speed-up, making it suitable for large-scale virtual screening.
• This approach holds potential for accelerating the identification of novel drug candidates and reducing drug discovery costs.