A neighborhood-regularization method leveraging multiview data for predicting the frequency of drug-side effects

Area of Science:

• Pharmacovigilance
• Computational Drug Safety
• Machine Learning in Healthcare

Background:

• Accurate drug benefit-risk assessment relies on determining drug side effect frequencies.
• Randomized controlled trials are the standard for this, but computationally predicting frequencies can guide these trials more effectively.
• Predicting drug side effect frequencies remains a significant computational challenge, with limited existing research.

Purpose of the Study:

• To propose and validate a novel computational method for predicting drug side effect frequencies.
• To leverage multiview data on drugs and side effects to improve prediction accuracy.
• To develop a method that can effectively guide randomized controlled trials by providing reliable side effect frequency estimates.

Main Methods:

• A neighborhood-regularization method (NRFSE) was developed, utilizing multiview data.
• Class-weighted non-negative matrix factorization decomposed the drug-side effect frequency matrix, employing Gaussian likelihood for unknown pairs.
• Multiview neighborhood regularization integrated drug and side effect attributes to ensure similar latent signatures for similar entities, adaptively weighting attributes.

Main Results:

• NRFSE demonstrated improved prediction performance compared to five state-of-the-art approaches on a benchmark dataset.
• Extensive experiments validated the method's effectiveness.
• An independent test set of post-marketing side effects further confirmed the efficacy of NRFSE in real-world scenarios.

Conclusions:

• The proposed NRFSE method offers a robust approach to predicting drug side effect frequencies.
• Leveraging multiview data and neighborhood regularization enhances prediction accuracy.
• NRFSE shows promise in aiding drug benefit-risk assessment and guiding clinical trial design.