Joint and individual variation explained (JIVE) for integrated analysis of multiple data types

Research in several fields now requires the analysis of data sets in which multiple high-dimensional types of data are available for a common set of objects. In particular, The Cancer Genome Atlas (TCGA) includes data from several diverse genomic technologies on the same cancerous tumor samples. In this paper we introduce Joint and Individual Variation Explained (JIVE), a general decomposition of variation for the integrated analysis of such data sets. The decomposition consists of three terms: a low-rank approximation capturing joint variation across data types, low-rank approximations for structured variation individual to each data type, and residual noise. JIVE quantifies the amount of joint variation between data types, reduces the dimensionality of the data and provides new directions for the visual exploration of joint and individual structures. The proposed method represents an extension of Principal Component Analysis and has clear advantages over popular two-block methods such as Canonical Correlation Analysis and Partial Least Squares. A JIVE analysis of gene expression and miRNA data on Glioblastoma Multiforme tumor samples reveals gene-miRNA associations and provides better characterization of tumor types. Data and software are available at https://genome.unc.edu/jive/

💡 Research Summary

The paper introduces Joint and Individual Variation Explained (JIVE), a statistical framework designed for the integrated analysis of multiple high‑dimensional data types measured on a common set of samples. Traditional methods such as Principal Component Analysis (PCA) handle only a single data matrix, while Canonical Correlation Analysis (CCA) and Partial Least Squares (PLS) focus on pairwise linear relationships and ignore structured variation that is unique to each data block. JIVE overcomes these limitations by decomposing each data block X_i (p_i × n) into three components: a shared low‑rank joint structure J_i, a low‑rank individual structure A_i specific to that block, and residual noise ε_i. The joint and individual components are constrained to have orthogonal row spaces, ensuring a unique and non‑overlapping representation of common and block‑specific variation.

Model ranks (r for the joint part and r_i for each individual part) are selected via a permutation‑based testing procedure, which assesses how much variance each candidate rank explains beyond chance. Once ranks are fixed, the algorithm iteratively minimizes the Frobenius norm of the residual matrix R = X – J – A by alternating between (1) updating each A_i given the current J (via truncated singular value decomposition of the residual X_i – J_i) and (2) updating J given the current A_i (via truncated SVD of the concatenated residual). The objective function decreases monotonically, guaranteeing convergence to a coordinate‑wise minimum.

The authors apply JIVE to The Cancer Genome Atlas (TCGA) Glioblastoma Multiforme (GBM) dataset, comprising 234 tumor samples with gene‑expression (23,293 variables) and miRNA‑expression (534 variables). After row‑centering and scaling each block to unit Frobenius norm, permutation testing (α = 0.01, 1,000 permutations) selects r = 5 joint components, r₁ = 33 gene‑specific components, and r₂ = 13 miRNA‑specific components. The joint structure accounts for 14 % of the variance in gene expression and 23 % in miRNA expression, reflecting biological processes that affect both data types. In contrast, 58 % of gene‑expression variance is captured by the individual component, indicating substantial gene‑specific variation unrelated to miRNA levels. Heatmaps of the estimated joint and individual matrices reveal coherent clusters of genes and miRNAs that share patterns across blocks, as well as distinct clusters that are unique to each block. These findings corroborate the known role of miRNAs as post‑transcriptional regulators while highlighting that many gene‑expression changes arise from other mechanisms.

Beyond this case study, JIVE is applicable to any number of data blocks, does not require the number of variables to exceed the number of samples, and provides a clear algebraic interpretation of the results. By quantifying the proportion of shared versus unique variation, JIVE offers a powerful exploratory tool for multi‑omics integration, enabling researchers to uncover biologically meaningful relationships that would be missed by analyzing each data type in isolation or by using methods that conflate joint and individual signals. The paper thus positions JIVE as a versatile extension of PCA, with distinct advantages over CCA and PLS for modern high‑dimensional, multi‑block data analysis.