A novel weighted pseudo-labeling framework based on matrix factorization for adverse drug reaction prediction

Area of Science:

• Pharmacovigilance
• Computational Drug Safety
• Machine Learning in Healthcare

Background:

• Adverse drug reactions (ADRs) pose significant global health risks and economic burdens.
• Accurate prediction of ADRs is vital for early intervention and patient safety.
• Current methods for predicting ADRs using drug-adverse reaction matrices suffer from data sparsity, limiting machine learning performance.

Purpose of the Study:

• To develop a novel weighted pseudo-labeling framework to address data sparsity in predicting drug-adverse reaction (ADR) associations.
• To enhance the classification performance of machine learning models for ADR prediction by integrating mined pseudo-labeled drug-ADR pairs.
• To improve the quality and prevent overfitting of pseudo-labels through a novel weighting approach.

Main Methods:

• Integration of multiple weighted matrix factorization (MF) models to mine potential unknown drug-ADR pairs.
• Utilizing a weighted pseudo-labeling strategy to incorporate newly identified drug-ADR pairs into the training set.
• Fine-tuning the MF model with pseudo-labeled data and employing a novel weighting approach for pseudo-labels to enhance robustness.

Main Results:

• The proposed weighted pseudo-labeling framework significantly outperformed baseline methods in predicting ADRs on sparse data from the SIDER database.
• Superior performance was demonstrated using Area Under Precision-Recall and F1-scores, particularly in increasingly sparse scenarios.
• A case study confirmed the framework's efficiency in real-world ADR prediction.

Conclusions:

• The novel weighted pseudo-labeling framework effectively mitigates data sparsity challenges in drug-ADR prediction.
• This approach enhances the accuracy and reliability of predicting adverse drug reactions, contributing to improved patient safety.
• The method shows promise for real-world applications in pharmacovigilance and drug safety monitoring.