TABLE VIII.

Example studies using bi-multivariate correlation methods, which aim to identify multi-SNP-multi-QT associations from high dimensional imaging genomic data.

Ref	Notes
[169]	S2CCA (structure aware SCCA), group 11 norm on both SNP and QT sides, SNPs grouped by LD block, QTs grouped by ROI
[108]	KG-SCCA (knowledge-guided SCCA), group l1 norm on genetic side (SNPs grouped by LD block), graph Laplacian type norm on imaging side (ROIs connected by co-expression network)
[171]	GNC-SCCA (generic non-convex penalty SCCA), seven non-convex penalties replacing l1 norm to reduce estimation bias
[172]	TLP-SCCA (truncated l1-norm penalized SCCA), TGL-SCCA (truncated group lasso SCCA), better approximation of l0 norm, voxels grouped by ROI, SNPs grouped by LD block
[174]	AGN-SCCA (absolute value based GraphNet SCCA), incorporation of a GraphNet variant into SCCA, joint selection of both positively and negatively correlated features
[175]	FDR-corrected SCCA, incorporation of FDR concept into SCCA
[177]	MTSCCA (multi-task SCCA), relating SNP to multimodal imaging QTs, l2,1 norm for SNP selection and QT selection
[178]	TG-SCCA (temporally constrained group SCCA), l1 for SNP selection, l2,1 for ROI selection (over time), fussed lasso for smoothing weights between neighbouring time points
[179]	T-MTSCCA (temporal multi-task SCCA), l1 and l2,1 for SNP and QT selection, fused pairwise l2,1 norm for smoothing weights between neighbouring time points
[180]	FSPLS (filtering + sparse Partial Least Square), two step procedure, univariate filtering, sparse PLS with l1 regularization
[181]	G-PDC (Greedy projected distance correlation), examination of pairwise gene-ROI associations, an efficient algorithm
[183]	DCCA (Distance CCA), identification of SNP set and QT set with the highest distance correlation
[186]	pICA (parallel independent component analysis), joint maximization of within-modality component independence and between-modality component correlation