Machine learning-guided dose-time optimization and experimental validation enhance coumarin therapeutics in oncology

Area of Science:

• Natural Products Chemistry
• Computational Biology
• Pharmacology

Background:

• Coumarins show promise as anticancer agents with targeted effects and low toxicity.
• Therapeutic use is hindered by data gaps and experimental inconsistencies.

Purpose of the Study:

• To apply machine learning (ML) regression models for predicting dose-exposure time windows for natural coumarins (esculetin, umbelliprenin, auraptene, galbanic acid) related to 50% cell viability.
• To validate ML predictions through in vitro experiments on a human colon carcinoma cell line.

Main Methods:

• Extracted in vitro data from 64 studies on coumarin dose, time, and cell viability.
• Employed eight ML regression algorithms (e.g., Gradient Boosting, Random Forest, SVM) to analyze dose-response relationships.
• Conducted in vitro validation experiments using auraptene and galbanic acid on LoVo cells.

Main Results:

• Esculetin, umbelliprenin, and auraptene demonstrated time-dependent anticancer activity.
• Galbanic acid showed cytotoxicity mainly at high concentrations and prolonged exposure.
• Gradient boosting ensemble models' predictions for auraptene and galbanic acid were validated by in vitro experiments.

Conclusions:

• ML models, particularly ensemble methods, show potential for predicting natural coumarin efficacy in oncology.
• This approach can help reduce experimental complexity in drug discovery.
• Further validation across diverse cell lines and coumarins is recommended.