SAFER: sub-hypergraph attention-based neural network for predicting effective responses to dose combinations

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Drug combinations offer significant benefits in cancer therapy but pose risks of increased toxicity.
• Current AI models for predicting drug synergy often overlook crucial dosage information and dynamic biological interactions.

Purpose of the Study:

• To develop an advanced AI model that accurately predicts drug combination synergy, considering dose-specific effects and dynamic biological networks.
• To overcome limitations of existing models that neglect dosage and static interaction data.

Main Methods:

• Introduction of SAFER (Sub-hypergraph Attention-based graph model), which incorporates complex biological relationships and dose-dependent effects.
• Utilizing subject-specific networks and attention mechanisms to model dynamic interactions.

Main Results:

• SAFER demonstrated superior performance compared to existing models on benchmark and independent datasets.
• Analysis revealed key biological pathways and genes, such as JAK-STAT signaling, implicated in lung cancer and fibrosis.

Conclusions:

• SAFER provides an interpretable framework for identifying drug-responsive signals and understanding dose-level combination effects.
• The model facilitates personalized medicine by enabling the prioritization of effective and safe drug combinations based on individual molecular profiles.