HINT: Hierarchical interaction network for clinical-trial-outcome predictions

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Development

Background:

• Clinical trial outcomes are often unpredictable due to safety, efficacy, and recruitment challenges.
• Accurate prediction of clinical trial success is vital for efficient drug development.

Purpose of the Study:

• To propose and evaluate the Hierarchical Interaction Network (HINT) for predicting clinical trial outcomes.
• To leverage multi-modal data and knowledge-embedding for improved prediction accuracy.

Main Methods:

• HINT encodes multi-modal data including drug molecules, target diseases, and eligibility criteria into embeddings.
• Knowledge-embedding modules are trained using drug pharmacokinetic and historical trial data.
• A hierarchical interaction graph integrates embeddings to predict trial outcomes.

Main Results:

• HINT achieved F1 scores of 0.665 (Phase I), 0.620 (Phase II), and 0.847 (Phase III) on independent test sets.
• The model demonstrated superior performance compared to the best baseline methods across most metrics.
• Validation was performed on a large dataset encompassing 1,160 Phase I, 4,449 Phase II, and 3,436 Phase III trials.

Conclusions:

• HINT offers a robust framework for predicting clinical trial outcomes by integrating diverse data sources.
• The developed model shows significant potential to enhance the efficiency and success rate of drug development pipelines.
• The benchmark dataset and code are publicly available to facilitate further research.