It is critical that the genome build (e.g. hg38) of the LD reference matches the genome build used to train the prediction models. You can use the function get_predictdb_genome_build() to check the genome build of PredictDB weight file.
The genome build of the GWAS summary statistics does not matter because variant positions are determined by the LD reference.
By default, we use our own LD matrices in “.RDS” format. We also support other common formats, such as “.Rdata”, “.mtx”, “.csv”, “.txt”. If you already have LD matrices in custom format, you could directly load your own LD matrices, and don’t need to save those in our LD matrix format. To do this, you could write a loader function that loads your LD matrices into our LD matrix format. Then set LD_format = "custom" and pass your loader function to the argument LD_loader_fun.
If your variant information files are custom format, you can also write a loader function to read your own variant information files, and then pass your loader function to the argument snpinfo_loader_fun.
cTWAS provides a function convert_geno_to_LD_matrix() to create LD matrices from individual level genotype data. See the tutorial “cTWAS utility functions” for details.
Running cTWAS with a different ancestry is under development. For now, we recommend using LD reference matching the GWAS samples (not the weights).
To save running time, we parallelized computations when running cTWAS main function and several other modular functions. We currently use the forking approach for parallelization, so it does not work on Windows.
You could set ncore to control how many cores are used in the parallelization. When processing LD matrices, especially during screening regions and fine-mapping, we recommend using larger memory. If you use the cTWAS main function, we have an argument ncore_LD, which allows using a lower number of cores during screening regions and fine-mapping steps, so more memory will be allocated for those steps.
The most important parameters of cTWAS are the “enrichment” parameters, the ratio of prior inclusion probabilities of molecular traits vs. variants. In our experience, these numbers usually range from 20 - 200. If the numbers are outside this range, check the prior variance parameters. When the variance parameters are very different between molecular traits and variants, you can get exceedingly large enrichment. In that case, you can specify the same prior variance parameters for molecular traits and genes, by setting group_prior_var_structure = "shared_all".
We limit variants in the processed weights to the processed variants in z_snp (by setting gwas_snp_ids = z_snp$id), so you should first run preprocess_z_snp(), and then run preprocess_weights() after getting processed variants. If you later changed the SNPs included in z_snp, you would need to rerun preprocess_weights() with gwas_snp_ids set to the updated z_snp$id.
This message suggests conflicts between the correlation matrices and the Z-scores when running fine-mapping. This often happens when you run cTWAS main function or fine-mapping using new data but load correlation matrices saved from an earlier analysis. By default, cTWAS will automatically load correlation matrices saved in cor_dir, so it will cause conflicts when you run different cTWAS analyses with the same cor_dir. So please make sure to save correlation matrices to a new directory when running new cTWAS analyses. Each cTWAS analysis should use a separate cor_dir to store the correlation matrices.
If you get an error message “Viewport has zero dimension(s)” when making locus plots, it usually suggests the R viewer window is too small. You can try increasing the size of the viewer window size and try again.
We included several default shapes to represent different types of molecular traits. If you have more types of molecular traits or if you want to change the shapes, you can set the point.shapesargument in the function make_locusplot().
This message is often caused by insufficient memory when running cTWAS with multiple cores in parallel. You could use a larger memory, or set a lower ncore. If you still get this error message, try ncore=1 to run without parallelization.