MVSL-DSF: Multiview Subspace Representation Learning and Cross-Modal Feature Dynamic Aggregation for Enhanced Drug...

Area of Science:

• Pharmacovigilance and Drug Safety
• Computational Pharmacology
• Machine Learning in Healthcare

Background:

• Drug side effects contribute significantly to morbidity and mortality.
• Accurate assessment of drug side effect frequency is vital for drug development and risk analysis.
• Current methods often use single-view graph neural networks, limiting the use of inter-drug and inter-side effect information and multivariate data features.

Purpose of the Study:

• To propose a novel multiview subspace representation learning aggregation method (MVSL-DSF) for predicting drug side effect frequencies.
• To overcome limitations of existing methods by integrating multivariate data features from various pharmacological networks.
• To enhance drug risk assessment and pharmacovigilance through precise quantification of drug side effect frequency.

Main Methods:

• Developed MVSL-DSF, a method mapping drug, side effect, and interaction views into a shared low-dimensional space.
• Employed Canberra-distance-driven cross-modal attention to optimize multiview consistency and complementarity.
• Sparsely aggregated features while eliminating redundant information to enhance model discriminative capability.

Main Results:

• MVSL-DSF achieved optimized Root Mean Square Error (RMSE) of 0.297 and Mean Absolute Error (MAE) of 0.167 in baseline experiments.
• Demonstrated superior performance compared to existing state-of-the-art methods.
• Constructed a frequency-severity matrix for predicting drug risk levels.

Conclusions:

• Multiview modeling enables precise quantification of drug side effect frequency from multisource data.
• The proposed MVSL-DSF method enhances representation, differentiation, and complementarity of features.
• Findings support improved risk assessment, clinical decisions, medication safety, and pharmacovigilance.