Preclinical side effect prediction through pathway engineering of protein interaction network models.

Area of Science:

• Pharmacology
• Bioinformatics
• Systems Biology

Background:

• Computational tools aim to predict drug side effects, but current protein-protein interaction (PPI) models exhibit limitations.
• PPI models often overpredict drug phenotypes and require precisely defined pathway phenotypes for accurate predictions.

Purpose of the Study:

• To evaluate and enhance the performance of PPI models, specifically PathFX, in predicting drug side effects.
• To develop improved pathway phenotype definitions using pathway engineering strategies.

Main Methods:

• Utilized PathFX, a PPI tool, to predict side effects for active ingredient-side effect pairs from drug labels.
• Developed novel pathway phenotypes through network-based and gene expression-based approaches (pathway engineering).
• Compared PPI model predictions against animal model data.

Main Results:

• Initial PPI model performance was limited, showing a trade-off between sensitivity and specificity.
• Pathway engineering strategies, including true positive examples and omics data, helped limit overprediction.
• Predictions from PPI models showed comparable performance metrics to animal models, indicating their utility despite imperfect evaluation metrics.

Conclusions:

• Pathway engineering is a promising strategy to improve the accuracy and utility of PPI network models for drug side effect prediction.
• PPI models can be valuable tools for drug effect prediction even without perfect evaluation metrics, especially when enhanced with pathway engineering.