From 1c3b93a12e3e0563ea8552ef5f120afffeb06c1b Mon Sep 17 00:00:00 2001
From: Gavin Rhys Lloyd <grlloyd@users.noreply.github.com>
Date: Thu, 20 Jul 2023 10:53:55 +0100
Subject: [PATCH] update d-ratio tests

---
 tests/testthat/test-dratio-filter.R | 121 ++++++++++++++++++++++++++--
 1 file changed, 114 insertions(+), 7 deletions(-)

diff --git a/tests/testthat/test-dratio-filter.R b/tests/testthat/test-dratio-filter.R
index d7b9197..7d9eb7c 100644
--- a/tests/testthat/test-dratio-filter.R
+++ b/tests/testthat/test-dratio-filter.R
@@ -1,12 +1,119 @@
 # filter d-ratio
-test_that('d-ratio filter works as expected',{
+test_that('d-ratio filter works as expected using sd ratio',{
     set.seed('57475')
     # DatasetExperiment
     DE = MTBLS79_DatasetExperiment(filtered=TRUE)
     M = dratio_filter(
         threshold=20,
         qc_label = 'QC',
-        factor_name='Class'
+        factor_name='Class',
+        method = 'ratio',
+        dispersion = 'sd'
+    )
+    M = model_apply(M,DE)
+    
+    # manually compute d-ratio
+    qc_mad=unlist(
+        lapply(DE$data[DE$sample_meta$Class=='QC',],sd,na.rm=TRUE)
+    )
+    sa_mad=unlist(
+        lapply(DE$data[DE$sample_meta$Class!='QC',],sd,na.rm=TRUE)
+    )
+    dr = (qc_mad/(sa_mad))*100
+    
+    # check manual vs fcn
+    expect_true(all(dr==M$d_ratio$d_ratio))
+    
+    # check values havent changed
+    expect_equal(dr[[1]],23.05762,tolerance = 5e-6)
+    expect_equal(dr[[100]],64.49242,tolerance = 5e-6)
+    
+    # just number of filtered columns
+    expect_true(ncol(predicted(M))==725)
+    expect_true(ncol(DE)-ncol(predicted(M))==854)
+    expect_true(ncol(DE)-ncol(predicted(M))==sum(dr>20))
+})
+
+test_that('d-ratio filter works as expected using mad ratio',{
+    set.seed('57475')
+    # DatasetExperiment
+    DE = MTBLS79_DatasetExperiment(filtered=TRUE)
+    M = dratio_filter(
+        threshold=20,
+        qc_label = 'QC',
+        factor_name='Class',
+        method = 'ratio',
+        dispersion = 'mad'
+    )
+    M = model_apply(M,DE)
+    
+    # manually compute d-ratio
+    qc_mad=unlist(
+        lapply(DE$data[DE$sample_meta$Class=='QC',],mad,na.rm=TRUE)
+    )
+    sa_mad=unlist(
+        lapply(DE$data[DE$sample_meta$Class!='QC',],mad,na.rm=TRUE)
+    )
+    dr = (qc_mad/(sa_mad))*100
+    
+    # check manual vs fcn
+    expect_true(all(dr==M$d_ratio$d_ratio))
+    
+    # check values havent changed
+    expect_equal(dr[[1]],14.14758,tolerance = 5e-6)
+    expect_equal(dr[[100]],48.23871,tolerance = 5e-6)
+    
+    # just number of filtered columns
+    expect_true(ncol(predicted(M))==834)
+    expect_true(ncol(DE)-ncol(predicted(M))==745)
+    expect_true(ncol(DE)-ncol(predicted(M))==sum(dr>20))
+})
+
+test_that('d-ratio filter works as expected using sd euclidean',{
+    set.seed('57475')
+    # DatasetExperiment
+    DE = MTBLS79_DatasetExperiment(filtered=TRUE)
+    M = dratio_filter(
+        threshold=20,
+        qc_label = 'QC',
+        factor_name='Class',
+        method = 'euclidean',
+        dispersion = 'sd'
+    )
+    M = model_apply(M,DE)
+    
+    # manually compute d-ratio
+    qc_mad=unlist(
+        lapply(DE$data[DE$sample_meta$Class=='QC',],sd,na.rm=TRUE)
+    )
+    sa_mad=unlist(
+        lapply(DE$data[DE$sample_meta$Class!='QC',],sd,na.rm=TRUE)
+    )
+    dr = qc_mad/sqrt((qc_mad^2) + (sa_mad^2))*100
+    
+    # check manual vs fcn
+    expect_true(all(dr==M$d_ratio$d_ratio))
+    
+    # check values havent changed
+    expect_equal(dr[[1]],22.46809,tolerance = 5e-6)
+    expect_equal(dr[[100]],54.19862,tolerance = 5e-6)
+    
+    # just number of filtered columns
+    expect_true(ncol(predicted(M))==747)
+    expect_true(ncol(DE)-ncol(predicted(M))==832)
+    expect_true(ncol(DE)-ncol(predicted(M))==sum(dr>20))
+})
+
+test_that('d-ratio filter works as expected using mad euclidean',{
+    set.seed('57475')
+    # DatasetExperiment
+    DE = MTBLS79_DatasetExperiment(filtered=TRUE)
+    M = dratio_filter(
+        threshold=20,
+        qc_label = 'QC',
+        factor_name='Class',
+        method = 'euclidean',
+        dispersion = 'mad'
     )
     M = model_apply(M,DE)
     
@@ -17,17 +124,17 @@ test_that('d-ratio filter works as expected',{
     sa_mad=unlist(
         lapply(DE$data[DE$sample_meta$Class!='QC',],mad,na.rm=TRUE)
     )
-    dr = (qc_mad/(qc_mad+sa_mad))*100
+    dr = qc_mad/sqrt((qc_mad^2) + (sa_mad^2))*100
     
     # check manual vs fcn
     expect_true(all(dr==M$d_ratio$d_ratio))
     
     # check values havent changed
-    expect_equal(dr[[1]],12.39,tolerance = 1e3)
-    expect_equal(dr[[100]],32.54,tolerance = 1e3)
+    expect_equal(dr[[1]],14.00808,tolerance = 5e-6)
+    expect_equal(dr[[100]],43.44776,tolerance = 5e-6)
     
     # just number of filtered columns
-    expect_true(ncol(predicted(M))==1052)
-    expect_true(ncol(DE)-ncol(predicted(M))==527)
+    expect_true(ncol(predicted(M))==850)
+    expect_true(ncol(DE)-ncol(predicted(M))==729)
     expect_true(ncol(DE)-ncol(predicted(M))==sum(dr>20))
 })