Biomedical datasets increasingly combine multiple cohorts and multiple data views, but most integrative methods focus on only one of those dimensions. The few bidimensional approaches are predominantly unsupervised and ignore covariates. In a preprint on arXiv, Jiuzhou Wang and Eric F. Lock introduce baRRR, a Bidimensional Augmented Reduced Rank Regression method that models both dimensions simultaneously.
The method estimates covariate-related and covariate-unrelated effects concurrently, while decomposing effect levels into global, partially shared, and individual components across cohorts and views. This allows the model to borrow strength across groups and modalities where appropriate, while still preserving group-specific structure.
In extensive simulations, baRRR showed advantages over existing approaches in signal recovery, outcome prediction, and missing-data imputation. The authors also applied the method to The Cancer Genome Atlas, using somatic mutations as covariates to inform patterns of variability across different omics platforms and cancer types. They position baRRR as a robust and adaptable tool for integrative biomedical analysis.