A guide for active learning in synergistic drug discovery.

Area of Science:

• Computational drug discovery
• Artificial intelligence in pharmacology
• Machine learning for drug synergy

Background:

• Synergistic drug combination screening is crucial for drug discovery but faces challenges due to a vast and complex search space.
• Current artificial intelligence (AI) models for predicting drug synergy are limited by the rarity of synergistic drug pairs in screening data.
• Active learning offers a potential solution by integrating experimental validation into the AI model's learning cycle.

Purpose of the Study:

• To investigate the key components and optimal implementation strategies for active learning in the context of drug synergy prediction.
• To evaluate the impact of different features, such as molecular encoding and cellular environment, on prediction performance.
• To quantify the efficiency gains of active learning in discovering synergistic drug combinations.

Main Methods:

• Exploration of active learning strategies for synergistic drug pair identification.
• Assessment of the influence of molecular encoding and cellular environment features on prediction accuracy.
• Analysis of active learning performance across different batch sizes and exploration-exploitation strategies.

Main Results:

• Molecular encoding had a minimal effect on prediction performance, whereas cellular environment features substantially improved synergy predictions.
• Active learning successfully identified 60% of synergistic drug pairs while exploring only 10% of the combinatorial chemical space.
• Smaller batch sizes further increased the synergy yield ratio, with dynamic tuning of exploration-exploitation strategies offering additional performance enhancements.

Conclusions:

• Active learning is a highly effective strategy for accelerating the discovery of synergistic drug combinations.
• Incorporating cellular environment features is critical for improving AI-driven synergy prediction models.
• Optimizing batch size and exploration strategies in active learning can significantly boost the efficiency of drug discovery pipelines.