AI-driven discovery of synergistic drug combinations against pancreatic cancer

Area of Science:

• Oncology
• Computational Biology
• Pharmacology

Background:

• Pancreatic cancer treatment frequently involves multi-drug regimens, but identifying optimal synergistic combinations remains a significant challenge.
• Existing methods for drug combination screening are often time-consuming and may not efficiently identify all potentially effective pairings.

Purpose of the Study:

• To evaluate predictive approaches, specifically machine learning (ML), for identifying synergistic drug combinations against pancreatic cancer.
• To experimentally validate predicted synergistic drug combinations and assess the performance of different ML models.

Main Methods:

• Screening 496 drug combinations of 32 anticancer compounds against PANC-1 pancreatic cancer cells to determine experimental synergy.
• Applying ML models, including graph convolutional networks and random forest, to predict synergy across 1.6 million potential drug combinations using the experimental data.
• Experimentally testing 88 predicted combinations to validate the ML models' predictions.

Main Results:

• Machine learning models successfully predicted synergistic drug combinations, with 51 out of 88 tested combinations showing synergy.
• Graph convolutional networks demonstrated the highest hit rate, while random forest achieved the highest precision in predicting synergy.
• The study identified 307 novel, experimentally validated synergistic drug combinations for pancreatic cancer treatment.

Conclusions:

• Machine learning offers a powerful and efficient strategy for discovering synergistic drug combinations in pancreatic cancer therapy.
• The validated synergistic combinations provide a valuable resource for the development of improved multi-drug treatment regimens.
• This research highlights the potential of predictive computational approaches to accelerate drug discovery and enhance cancer treatment strategies.