Leveraging pharmacovigilance data to predict population-scale toxicity profiles of checkpoint inhibitor immunotherapy

Area of Science:

• Oncology
• Computational Biology
• Pharmacovigilance

Background:

• Immune checkpoint inhibitors (ICIs) have revolutionized cancer immunotherapy but are associated with complex toxicities.
• Predicting and managing these toxicities remains a significant challenge for clinical outcomes and computational analysis.

Purpose of the Study:

• To introduce DySPred, a deep learning framework for predicting ICI-induced toxicity profiles at the population level.
• To leverage real-world pharmacovigilance data for accurate toxicity risk assessment across diverse patient groups and cancer types.

Main Methods:

• Development of DySPred, a dynamic graph convolutional network-based deep learning framework.
• Utilizing large-scale real-world pharmacovigilance data for model training and validation.
• Assessing the alignment of drug-induced transcriptional alterations with toxicity profiles for small-molecule antineoplastic agents.

Main Results:

• DySPred accurately predicts ICI toxicity risks across various demographic cohorts and cancer types.
• The framework demonstrates resilience in scenarios with limited sample sizes.
• DySPred reveals temporal trends in ICI-induced toxicities and aligns toxicity profiles with transcriptional data.

Conclusions:

• DySPred offers a versatile methodology for population-level profiling of ICI-induced toxicities.
• The framework enables proactive toxicity monitoring and timely treatment adjustments in cancer immunotherapy.
• This approach supports the advancement of safer and more effective cancer immunotherapy strategies.