pdCSM-cancer: Using Graph-Based Signatures to Identify Small Molecules with Anticancer Properties

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Developing new anticancer drugs is challenging due to low hit rates.
• Existing predictive models have limited success in identifying broad-spectrum anticancer compounds.
• There is a need for advanced tools to predict molecule activity against diverse cancer cell lines.

Purpose of the Study:

• To develop a novel predictive tool, pdCSM-cancer, for accurately identifying molecules with anticancer properties.
• To create a comprehensive platform for predicting anticancer bioactivity based on chemical structure.
• To enhance the efficiency of anticancer drug discovery and screening processes.

Main Methods:

• Utilized a graph-based signature representation of small molecule chemical structures.
• Developed and validated predictive models on experimental growth inhibition concentration (GI50%) data.
• Trained models on over 18,000 compounds across 9 tumor types and 74 cancer cell lines.
• Created a generic model to predict activity across at least 60 cell lines.

Main Results:

• Achieved high performance with Pearson's correlation coefficients up to 0.74 in cross-validation and 0.67 in blind tests.
• The generic model demonstrated strong predictive power with an area under the receiver operating characteristic curve (AUC) up to 0.94.
• The tool, pdCSM-cancer, outperforms alternative prediction approaches.

Conclusions:

• pdCSM-cancer is a comprehensive and accurate platform for predicting anticancer molecule activity.
• The tool can significantly aid in optimizing screening libraries and identifying effective anticancer drug candidates.
• pdCSM-cancer is freely available online, providing a valuable resource for cancer research.