Subset-binding: A novel algorithm to detect paired itemsets from heterogeneous data including biological datasets

The integration of heterogeneous data to infer latent relationships across them and find the factors in the relationship is a challenging task. In this regard, various machine learning techniques have provided novel insights through data integration. However, concerns remain regarding their application to biological datasets because the latent consensus information across all views is often limited to partial components that do not have a significant impact on the mutual agreement across views. Advocating the idea of “subset-binding,” which focuses on finding inter-related attributes in heterogeneous data according to their co-occurrence, this study developed a novel algorithm to perform subset-binding by extending fuzzy association rule mining techniques. Our method could detect genes related to liver toxicity caused by acetaminophen in a data-driven manner; the results are consistent with those reported in the literature. This technology paves the way for a wide range of applications, including biomarker detection and patient stratification.