updated docs and sim12

mkdocs/usage.md

@@ -3,9 +3,9 @@
 ```shell
 ./varGWAS
 
-Program to perform vGWAS of trait against variants in the BGEN format
+Program to perform GWAS of trait variability against variants in the BGEN format
 Usage:
-  varGWAS C++ v1.0.0 [OPTION...]
+  varGWAS v1.2.1 [OPTION...]
 
   -v, --variable_file arg  Path to phenotype file
   -s, --sep arg            File separator
@@ -14,45 +14,19 @@ Usage:
   -b, --bgen_file arg      Path to BGEN file
   -p, --phenotype arg      Column name for phenotype
   -i, --id arg             Column name for genotype identifier
-  -r, --robust             Robust method using median value (Brown-Forsythe)
+  -m, --maf arg            Filter out variants with a MAF below this threshold
   -h, --help               Print usage
-  -t, --threads arg        Number of threads (default: 8)
+  -t, --threads arg        Number of threads
 ```
 
 - Unordered categorical variables should be one-hot encoded.
 - Do not provide null values in the phenotype file - these should be filtered out.
 
-# Docker
+# Covariates
 
-Perform GWAS
+In addition to standard covariates, also include the square of continuous/ordinal phenotypes to adjust the variance effect. 
 
-```shell
-docker run \
--v /Users/ml18692/projects/varGWAS/test/data:/data \
--e SPDLOG_LEVEL=debug \
--it vargwas \
--v /data/phenotypes.csv \
--s , \
--c sex,age,PC.1,PC.2,PC.3,PC.4,PC.5,PC.6,PC.7,PC.8,PC.9,PC.10 \
--o /data/output.txt \
--b /data/genotypes.bgen \
--p Y \
--i S \
--t 1
-```
-
-# Unit tests
-
-Run unit tests
-
-```shell
-mkdir -p build
-cd build
-cmake .. -DCMAKE_BUILD_TYPE=Debug
-make
-```
-
-# Simulation
+# Simulations
 
 See [README](https://github.com/MRCIEU/varGWAS/blob/master/sim/README.md)
 
@@ -65,3 +39,17 @@ the [spdlog](https://github.com/gabime/spdlog#load-log-levels-from-env-variable-
 export SPDLOG_LEVEL=debug
 ./varGWAS
 ```
+
+# Unit tests
+
+Run unit tests
+
+```shell
+# build debug release
+mkdir -p build
+cd build
+cmake .. -DCMAKE_BUILD_TYPE=Debug
+make
+# run tests
+./bin/varGWAS_test
+```

sim/README.md

@@ -121,10 +121,10 @@ for p in $(seq 0 0.5 6); do
     sbatch runR.sh sim12.R -p "$p" -i <i>
     cd ..
 done
-head -n1 0/sim12_0.csv > results.csv
-cat */sim12_*.csv | grep -v b0_dummy >> results.csv
-head -n1 0/sim12_0.csv > results.csv
-cat */sim12_*.csv | grep -v b0_dummy >> results.csv
+head -n1 0.0/sim12_0_0.csv > results.csv
+cat */sim12_*csv | grep -v b1_dummy >> results.csv
+head -n1 0.0/sim12_0_1.csv > results.csv
+cat */sim12_*csv | grep -v b1_dummy >> results.csv
 ```
 
 ## Sim13 - Adjusting the variance effect for the interaction

sim/sim12.R

@@ -18,8 +18,8 @@ n_obs <- 10000
 n_sim <- 1000
 af <- 0.4
 
-# main effect size of X on Y explaining 5% variance
-delta <- 0.33
+# main effect size of X on Y detectable with 95% CI
+delta <- 0.0525
 theta <- delta * opt$phi
 
 # simulate GxE interaction effects and estimate power

sim/sim12_plot.R

@@ -62,8 +62,8 @@ dev.off()
 # coverage
 results <- merge(results, v1_mean, "phi")
 results <- merge(results, v2_mean, "phi")
-names(results)[17] <- "v1_mean"
-names(results)[18] <- "v2_mean"
+names(results)[15] <- "v1_mean"
+names(results)[16] <- "v2_mean"
 results$b1_dummy_lci <- results$b1_dummy - (1.96 * results$s1_dummy)
 results$b1_dummy_uci <- results$b1_dummy + (1.96 * results$s1_dummy)
 results$b2_dummy_lci <- results$b2_dummy - (1.96 * results$s2_dummy)

sim/sim1_plot.R

@@ -41,81 +41,49 @@ t <- fread(paste0("t/data/power_t.csv"))
 # process data
 
 # N
-cpp_bp_n <- calc_power(n, "P.cpp_bp", 200, c("phi", "lambda"))
-cpp_bp_n$dist <- "Normal"
-cpp_bp_n$method <- "Breusch-Pagan"
-
 cpp_bf_n <- calc_power(n, "P.cpp_bf", 200, c("phi", "lambda"))
 cpp_bf_n$dist <- "Normal"
 cpp_bf_n$method <- "Brown-Forsythe (LAD)"
 
-osca_levene_n <- calc_power(n, "P.osca_mean", 200, c("phi", "lambda"))
-osca_levene_n$dist <- "Normal"
-osca_levene_n$method <- "Levene"
-
 osca_bf_n <- calc_power(n, "P.osca_median", 200, c("phi", "lambda"))
 osca_bf_n$dist <- "Normal"
 osca_bf_n$method <- "Brown-Forsythe"
 
 # Mixed N
-cpp_bp_mn <- calc_power(mn, "P.cpp_bp", 200, c("phi", "lambda"))
-cpp_bp_mn$dist <- "Mixed normal"
-cpp_bp_mn$method <- "Breusch-Pagan"
-
 cpp_bf_mn <- calc_power(mn, "P.cpp_bf", 200, c("phi", "lambda"))
 cpp_bf_mn$dist <- "Mixed normal"
 cpp_bf_mn$method <- "Brown-Forsythe (LAD)"
 
-osca_levene_mn <- calc_power(mn, "P.osca_mean", 200, c("phi", "lambda"))
-osca_levene_mn$dist <- "Mixed normal"
-osca_levene_mn$method <- "Levene"
-
 osca_bf_mn <- calc_power(mn, "P.osca_median", 200, c("phi", "lambda"))
 osca_bf_mn$dist <- "Mixed normal"
 osca_bf_mn$method <- "Brown-Forsythe"
 
 # Lognormal
-cpp_bp_l <- calc_power(l, "P.cpp_bp", 200, c("phi", "lambda"))
-cpp_bp_l$dist <- "Lognormal"
-cpp_bp_l$method <- "Breusch-Pagan"
-
 cpp_bf_l <- calc_power(l, "P.cpp_bf", 200, c("phi", "lambda"))
 cpp_bf_l$dist <- "Lognormal"
 cpp_bf_l$method <- "Brown-Forsythe (LAD)"
 
-osca_levene_l <- calc_power(l, "P.osca_mean", 200, c("phi", "lambda"))
-osca_levene_l$dist <- "Lognormal"
-osca_levene_l$method <- "Levene"
-
 osca_bf_l <- calc_power(l, "P.osca_median", 200, c("phi", "lambda"))
 osca_bf_l$dist <- "Lognormal"
 osca_bf_l$method <- "Brown-Forsythe"
 
 # T-dist
-cpp_bp_t <- calc_power(t, "P.cpp_bp", 200, c("phi", "lambda"))
-cpp_bp_t$dist <- "T-dist"
-cpp_bp_t$method <- "Breusch-Pagan"
-
 cpp_bf_t <- calc_power(t, "P.cpp_bf", 200, c("phi", "lambda"))
 cpp_bf_t$dist <- "T-dist"
 cpp_bf_t$method <- "Brown-Forsythe (LAD)"
 
-osca_levene_t <- calc_power(t, "P.osca_mean", 200, c("phi", "lambda"))
-osca_levene_t$dist <- "T-dist"
-osca_levene_t$method <- "Levene"
-
 osca_bf_t <- calc_power(t, "P.osca_median", 200, c("phi", "lambda"))
 osca_bf_t$dist <- "T-dist"
 osca_bf_t$method <- "Brown-Forsythe"
 
 # combine
 results <- rbind(
-    cpp_bp_n, cpp_bf_n, osca_levene_n, osca_bf_n,
-    cpp_bp_mn, cpp_bf_mn, osca_levene_mn, osca_bf_mn,
-    cpp_bp_l, cpp_bf_l, osca_levene_l, osca_bf_l,
-    cpp_bp_t, cpp_bf_t, osca_levene_t, osca_bf_t
+    cpp_bf_n,  osca_bf_n,
+    cpp_bf_mn,  osca_bf_mn,
+    cpp_bf_l,  osca_bf_l,
+    cpp_bf_t,  osca_bf_t
 )
-results$method <- factor(results$method, levels = c("Breusch-Pagan", "Brown-Forsythe (LAD)", "Levene", "Brown-Forsythe"))
+results$method <- factor(results$method, levels = c("Brown-Forsythe", "Brown-Forsythe (LAD)"))
 results$dist <- factor(results$dist, levels = c("Normal", "Mixed normal", "Lognormal", "T-dist"))
 results$lambda <- factor(results$lambda, levels = c(1,10,100,1000))
 

sim/sim3.R

@@ -110,6 +110,7 @@ p1 <- qqgplot(results, "log", "P.cpp_bf")
 p2 <- qqgplot(results, "sqrt", "P.cpp_bf")
 p3 <- qqgplot(results, "irnt", "P.cpp_bf")
 p4 <- qqgplot(results, "cube_root", "P.cpp_bf")
+p5 <- qqgplot(results, "none", "P.cpp_bf")
 
 p <- ggarrange(p1, p2, p3, p4, labels = c("A", "B", "C", "D"), ncol = 2, nrow = 2)
 pdf("data/trans_t1e_bf.pdf")
@@ -120,6 +121,7 @@ p1 <- qqgplot(results, "log", "P.osca_median")
 p2 <- qqgplot(results, "sqrt", "P.osca_median")
 p3 <- qqgplot(results, "irnt", "P.osca_median")
 p4 <- qqgplot(results, "cube_root", "P.osca_median")
+p5 <- qqgplot(results, "none", "P.osca_median")
 
 p <- ggarrange(p1, p2, p3, p4, labels = c("A", "B", "C", "D"), ncol = 2, nrow = 2)
 pdf("data/trans_t1e_osca_median.pdf")