README.md

In ZhaotongL/GraphMRcML:

Brief introduction to using GraphMRcML

Step 1. Data preparation

First, to analyze the relationship among $N$ traits, you need to prepare your pre-processed data, including dat: a list of 5 elements: * The first two elements are b_mat and se_mat, which are both $m\times n$ matrices storing the GWAS estimates (and corresponding standard errors) of $M$ SNPs on $N$ traits. * The third element n_vec is a vector of length $N$, storing the sample size for each trait. * The fourth element P is an $N\times N$ correlation matrix, storing the correlations between the GWAS estimates among the $N$ traits, which can be obtained by, for example, bivariate LDSC regression. * The fifth element R_list is a list* of LD information among the $M$ SNPs, which is splitted into LD blocks. Each element in R_list has two elements, snp is a vector of SNP names (rsid), and R is the LD correlation matrix of SNPs in snp.

• IV_list: a list of $N\times (N-1)/2$ elements, which correspond to a total of $N\times (N-1)/2$ pairwise MR analyses among $N$ traits. Specifically, each element in IV_list stores the union of SNPs (rsid) to be used in the bivariate MR analysis for each pair of traits.

The sample codes for preprocessing the data are given in real_data/real_data.R and real_data/process.R.

Briefly, in real_data.R, we obtain IVs for each pair of traits by searching GWAS significant SNPs with trait A and GWAS significant SNPs with trait B, and do LD-clumping on the union. Then this LD-clumped union of SNPs is stored in IV_list. We then extract all $M$ SNPs in IV_list for all $N$ traits and perform data harmonizing.

In process.R, we construct the LD correlation matrix among the $M$ SNPs by first splitting them into LD blocks, and call ieugwasr::ld_matrix to obtain LD matrix.

Note 1. Some elements in b_mat and se_mat can be NA, as long as those SNPs will not be used in the bivariate analyses involving that trait, which is guaranteed by the use of IV_list. 2. Based on the preprocessing code, we expect the order in IV_list should be pair 1-2, pair 1-3, ..., pair 1-N, pair 2-3, pair 2-4, ..., pair (N-1)-N.

Step 2. Run GraphMRcML

After preparing the above two datasets, we can implement GraphMRcML. As GraphMRcML relies on data perturbation (bootstrap) for statistical inference, we can run the analysis in parallel. real_data/Panalyze.R shows an example to run the analysis with 10 perturbations (num_pert = 10) in each job, then you can submit, e.g. 200 jobs, to obtain a total of 2000 perturbations.

Step 3. Summarize and visualize results

After Step 2, you first need to combine results from all 200 jobs (see real_data/combine_list.R), then you can summarize the result using subset_Graph_d1 and visualize it with plot_graph in R/helper.R. In particular, subset_Graph_d1 returns a list storing the final total-effect graph (obs_graph_mean,obs_graph_sd,obs_graph_pval) and the final direct-effect graph after network deconvolution (dir_graph_mean,dir_graph_sd,dir_graph_pval).

It also stores warn_len which is the number of direct-effect graphs with a spectral value greater than 1 (which violates the key assumption in ND), and conv_len which is the number of failed convergence of the algorithm to ensure the diagonal elements of the resulting direct-effect network are zeros. In general, if the two numbers are large, the resulting direct-effect graph may not be reliable. One strategy is to reduce the number of traits and/or highly correlated traits in the model.

Reference

Lin, Z., Xue, H., & Pan, W. (2023). Combining Mendelian randomization and network deconvolution for inference of causal networks with GWAS summary data. PLoS genetics, 19(5), e1010762.

ZhaotongL/GraphMRcML documentation built on May 31, 2024, 6:52 a.m.