Multiple Kronecker RLS fusion-based link propagation for drug-side effect prediction

TL;DR

This work tackles drug–side-effect prediction as a multi-view link prediction problem. It introduces MKronRLSF-LP, which learns a consensus partition across multiple Kron-RLS submodels and applies multiple graph Laplacian regularization to fuse views robustly without heavy explicit pairwise-matrix computations. The method constructs diverse drug and side-effect kernels, optimizes view weights and kernel coefficients in an iterative framework, and demonstrates superior AUPR and F-score across four real datasets relative to strong baselines. The results suggest that consensus-driven, Laplacian-regularized fusion yields more reliable predictions in sparse bipartite networks with heterogeneous information.

Abstract

Drug-side effect prediction has become an essential area of research in the field of pharmacology. As the use of medications continues to rise, so does the importance of understanding and mitigating the potential risks associated with them. At present, researchers have turned to data-driven methods to predict drug-side effects. Drug-side effect prediction is a link prediction problem, and the related data can be described from various perspectives. To process these kinds of data, a multi-view method, called Multiple Kronecker RLS fusion-based link propagation (MKronRLSF-LP), is proposed. MKronRLSF-LP extends the Kron-RLS by finding the consensus partitions and multiple graph Laplacian constraints in the multi-view setting. Both of these multi-view settings contribute to a higher quality result. Extensive experiments have been conducted on drug-side effect datasets, and our empirical results provide evidence that our approach is effective and robust.

Figures (6)

• Figure 1: Taxonomy of multi-view learning framework literature. Note: "Partition" commonly refers to the learned result. This concept is more commonly found in classification and clustering tasks liu2023contrastivebruno2009multiviewwang2019multi. (a) Early fusion: the views are combined before the training process is performed; (b) Late fusion: a different model for each view is separately trained and then a combination is taken as the final partition; (c) Fusion during the training phase: it has some degree of freedom to model the views differently but to also ensure that information from other views is exploited during the training phase.
• Figure 2: Framework diagram of MKronRLSF-LP. MKronRLSF-LP allow the multiple partitions have a degree of freedom to model the single information and introduce a multiple graph Laplacian regularization into consensus partition.
• Figure 3: Critical difference diagram of average score ranks. A crossbar is over each group of methods that do not show a statistically significant difference among themselves.
• Figure 4: Ablation study of the consensus partition and multiple graph Laplacian constraint on four datasets.
• Figure 5: Visualization of drug-side effect association problems.