Machine learning and deep learning-based drug-drug interactions prediction: a systematic review focused on anticancer...

Area of Science:

• Oncology
• Pharmacology
• Computational Biology

Background:

• Cancer patients frequently experience drug-drug interactions (DDIs) due to polypharmacy, increasing toxicity and reducing treatment efficacy.
• Traditional DDI identification methods (in vitro, in vivo, clinical studies, post-marketing surveillance) are often time-consuming and costly.
• Machine Learning (ML) and Deep Learning (DL) offer accelerated DDI prediction, crucial for managing complex oncology drug regimens.

Purpose of the Study:

• To systematically review ML- and DL-based DDI prediction models specifically involving anticancer drugs.
• To summarize key features of anticancer drugs, prediction model details (tasks, performance), and newly predicted DDIs.
• To assess the practical value and clinical relevance of computational DDI prediction in oncology.

Main Methods:

• Systematic literature review of ML/DL models for anticancer drug-drug interaction prediction.
• Analysis of prediction tasks (DDI existence/type) and model performance metrics.
• Verification of newly predicted DDIs using DrugBank and Drugs.com databases.

Main Results:

• A comprehensive summary of ML/DL models applied to anticancer DDIs was compiled.
• Newly predicted potential DDIs were identified, with 22 out of 96 pairs experimentally confirmed.
• The study demonstrated the practical utility of computational methods in identifying clinically relevant DDIs.

Conclusions:

• ML and DL models show significant promise for accelerating DDI identification in oncology.
• Computational approaches can uncover novel DDIs, complementing traditional methods.
• Tailoring novel prediction strategies to oncology drug characteristics can enhance clinical applicability and patient safety.