FlashfmZero is a computationally efficient approach to jointly fine-map signals in any number of uncorrelated quantitative traits (i.e. zero correlation), as may result from latent traits estimated from factor analysis using a varimax rotation. For correlated traits, the original flashfm multi-trait fine-mapping method should be used and cited, which is available in this package for convenience. Flashfm and flashfmZero output trait-specific results,leveraging information between traits; for each trait, credible sets, SNP marginal posterior probabilities of causality (MPP), and multi-SNP model posterior probabilities (PP) are output.
We also provide an approach to estimate latent factor GWAS summary statistics using only summary level data - observed trait GWAS summary statistics, observed trait covariance matrix.
For more details, please see:
F Zhou, WJ Astle, AS Butterworth, JL Asimit. (2024). Improved genetic discovery and fine-mapping resolution through multivariate latent factor analysis of high-dimensional traits. bioRxiv
Website available at: https://jennasimit.github.io/flashfmZero/
Example annotated scripts that only use observed trait summary-level data in our analysis of the INTERVAL study’s NMR metabolic traits (Karjalainen et al. 2024) are available on the flashfmZero website in the article tabs. These scripts (with minor modifications) were also used in our analysis of 99 blood cell traits from the INTERVAL study (Astle et al. 2016, Akbari et al. 2023).
See Estimation of latent factors and Latent factor GWAS and fine-mapping for annotated example R scripts that use summary-level data and are broadly applicable to other GWAS data.
For proof of principle, we have applied these methods to GWAS results from 99 raw blood cell traits and their 25 latent factors, derived from complete individual-level data, in the INTERVAL study. These analysis scripts are available here: https://github.com/fz-cambridge/flashfmZero-INTERVAL-analysis
flashfmZero could be installed with ease on versions of R > 4.2.1 and is compatible with all platforms.
Installation time is estimated as 2 minutes.
Specific requirements for Windows and Mac platforms follow.
Windows
Must install Rtools.
Mac
Must have the following installed (details at R for MacOS):
-
Xcode: free on the Apple App Store
-
Fortran compiler. R 4.3.0 and higher use universal GNU Fortran 12.2 compiler and an installer package is available here: gfortran-12.2-universal.pkg (242MB)
# install.packages("devtools")
devtools::install_github("jennasimit/flashfmZero")The following packages from CRAN and Bioconductor are required:
install.packages("parallel")
install.packages("Matrix")
install.packages("gtools")
install.packages("rlist")NB: Must have a Java JDK installed in order to run R2BGLiMS (R2BGLiMS is pre-installed in the flashfmZero package, so no need to install it separately). This is only needed if you need to run single-trait fine-mapping using JAM. If single-trait fine-mapping results are available, then it is not necessary to have Java JDK installed.
remotes::install_github("jennasimit/flashfmZero")
library(flashfmZero)