FDR improves results

banskt · Jan 7, 2019 · 2e5d774 · 2e5d774
1 parent 11106e5
commit 2e5d774
Show file tree

Hide file tree

Showing 5 changed files with 93 additions and 26 deletions.
diff --git a/analysis/jupyter/crosstissue_validation.ipynb b/analysis/jupyter/crosstissue_validation.ipynb
diff --git a/analysis/plots/cross_tissue_validation_ms_hlv_lmcorrected_sb0_01.pdf b/analysis/plots/cross_tissue_validation_ms_hlv_lmcorrected_sb0_01.pdf
diff --git a/analysis/plots/gtex_ms_hlv_lmcorrected_validation_tejaas-sb0_01.pdf b/analysis/plots/gtex_ms_hlv_lmcorrected_validation_tejaas-sb0_01.pdf
diff --git a/analysis/utils/load_results.py b/analysis/utils/load_results.py
@@ -51,6 +51,18 @@ def jpa(filepath):
 
 
 def matrixeqtl(filepath):
+    res = dict()
+    with open(filepath, 'r') as mfile:
+        next(mfile)
+        for line in mfile:
+            arr  = line.strip().split("\t")
+            rsid = arr[0]
+            pval = float(arr[4])
+            if rsid not in res:
+                res[rsid] = -np.log10(pval)
+    return res
+
+def matrixeqtl_fdr(filepath):
     res = dict()
     with open(filepath, 'r') as mfile:
         next(mfile)

diff --git a/analysis/utils/precision_recall_scores.py b/analysis/utils/precision_recall_scores.py
@@ -34,10 +34,10 @@ def confusion_matrix(result):
 
     #tpr    = np.array([x / pos if pos > 0 else 0 for x in tplist]) 				# TPR = Recall = TP / Positives
     #fpr    = np.array([x / neg if neg > 0 else 0 for x in fplist]) 				# FPR = FP / Negatives
-    #ppv    = np.array([x[0] / sum(x) if sum(x) > 0 else 1 for x in zip(tplist, fplist)]) 	# PPV = Precision = TP / (TP + FP)
+    ppv    = np.array([x[0] / sum(x) if sum(x) > 0 else 1 for x in zip(tplist, fplist)]) 	# PPV = Precision = TP / (TP + FP)
     #fdr    = np.array([x[1] / sum(x) if sum(x) > 0 else 0 for x in zip(tplist, fplist)]) 	# FDR = FP / (TP + FP)
     tparr  = np.array(tplist)                                                                   # Number of true positives
     nsel   = np.array([sum(x) for x in zip(tplist, fplist)]) 					# Number of y selected at each threshold
 
     #return fpr, tpr, ppv, nsel, fdr
-    return tparr, nsel
+    return tparr, nsel, ppv