DeepCancerMap: A versatile deep learning platform for target- and cell-based anticancer drug discovery.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Anticancer drug discovery faces challenges with time-consuming and costly experimental screening methods.
• Traditional target- and phenotypic-based screening approaches are labor-intensive and expensive.

Purpose of the Study:

• To develop a computational platform for efficient anticancer drug discovery.
• To predict the inhibitory activity of compounds against cancer targets and cell lines.
• To accelerate the identification and repositioning of potential anticancer agents.

Main Methods:

• Collected a large dataset of 485,900 compounds and 3,919,974 bioactivity records.
• Constructed 832 classification models (426 target-based, 406 cell-based) using the FP-GNN deep learning method.
• Developed the user-friendly webserver DeepCancerMap for various drug discovery tasks.

Main Results:

• FP-GNN models demonstrated superior predictive performance compared to classical and other deep learning methods.
• Achieved high AUC values of 0.91 (targets), 0.88 (academia-sourced cell lines), and 0.91 (NCI-60 cell lines).
• The DeepCancerMap platform enables virtual screening, profiling prediction, target fishing, and drug repositioning.

Conclusions:

• DeepCancerMap provides a powerful, accessible tool to accelerate anticancer drug discovery.
• The platform's high-quality predictive models facilitate efficient identification of novel therapeutic candidates.
• This computational approach significantly reduces the time and cost associated with traditional drug discovery pipelines.