remove genes with zero variance in specific batches

combat/pycombat.py

@@ -420,6 +420,27 @@ def check_NAs(dat):
     return(NAs)
 
 
+def remove_zero_variance_genes(dat, batches):
+    """Remove rows (genes) with zero variance in at least one batch
+
+    Arguments:
+        dat {matrix} -- the data matrix
+        batches {array} -- batch indices
+
+    Returns:
+        non_zero_variance_data {matrix} -- the data matrix without rows with zero variance in a batch
+        genes_to_remove {array} -- array of indices of genes to remove
+    """
+    genes_to_remove = []
+    for batch in batches:
+        batch_data = dat[:, batch]
+        zero_variance_index = batch_data.var(axis=1) == 0
+        genes_to_remove.append(zero_variance_index)
+    genes_to_remove = np.array(genes_to_remove).any(axis=0)
+    non_zero_variance_data = dat[~genes_to_remove,:]
+    return non_zero_variance_data, genes_to_remove
+
+
 def calculate_mean_var(design, batches, ref, dat, NAs, ref_batch, n_batches, n_batch, n_array):
     """ calculates the Normalisation factors
 
@@ -642,7 +663,7 @@ def pycombat(data, batch, mod=[], par_prior=True, prior_plots=False, mean_only=F
 
     list_samples = data.columns
     list_genes = data.index
-    dat = data.values
+    dat = data.copy().values
 
     check_mean_only(mean_only)
 
@@ -651,6 +672,9 @@ def pycombat(data, batch, mod=[], par_prior=True, prior_plots=False, mean_only=F
     n_batch, batches, n_batches, n_array = treat_batches(batch)
     design = treat_covariates(batchmod, mod, ref, n_batch)
     NAs = check_NAs(dat)
+    dat, genes_to_remove = remove_zero_variance_genes(dat, batches)
+    genes_to_keep = list_genes[~genes_to_remove]
+    genes_to_remove = list_genes[genes_to_remove]
     if not(NAs):
         B_hat, grand_mean, var_pooled = calculate_mean_var(
             design, batches, ref, dat, NAs, ref_batch, n_batches, n_batch, n_array)
@@ -662,9 +686,10 @@ def pycombat(data, batch, mod=[], par_prior=True, prior_plots=False, mean_only=F
         bayes_data = adjust_data(s_data, gamma_star, delta_star, batch_design,
                                 n_batches, var_pooled, stand_mean, n_array, ref_batch, ref, batches, dat)
 
-        bayes_data_df = pd.DataFrame(bayes_data,
+        reduced_bayes_data_df = pd.DataFrame(bayes_data,
                     columns = list_samples,
-                    index = list_genes)
+                    index = genes_to_keep)
+        bayes_data_df = pd.concat([reduced_bayes_data_df, data.loc[genes_to_remove]], axis=0).loc[data.index]
 
         return(bayes_data_df)
     else:

combat/test_unit.py

@@ -40,6 +40,7 @@
 from .pycombat import check_mean_only, define_batchmod, check_ref_batch, treat_batches, treat_covariates, check_NAs
 from .pycombat import calculate_mean_var, calculate_stand_mean
 from .pycombat import standardise_data, fit_model, adjust_data
+from .pycombat import remove_zero_variance_genes
 from .pycombat import pycombat
 
 ##########
@@ -140,6 +141,20 @@ def test_all_1():
     assert all_1(np.array([1.5,0.5,1,1,1])) == False # This test to show the limit of the method we use
 
 
+# tests for remove_zero_variance_genes function
+def test_remove_zero_variance_genes():
+    batches = [np.array([0,1,2,3]), np.array([4,5,6]), np.array([7,8])]
+    dat = np.array([[0,0,0,0,1,2,1,3,4],
+                    [0,1,2,3,0,1,2,1,2],
+                    [0,1,1,2,2,3,2,1,2],
+                    [1,2,2,1,0,0,0,1,1]])
+    reduced_dat, genes_to_remove = remove_zero_variance_genes(dat, batches)
+    print(genes_to_remove)
+
+    assert reduced_dat.shape == (2, 9)
+    assert all([a == b for a, b in zip(genes_to_remove, np.array([ True, False, False,  True]))])
+
+
 # test for check_mean_only
 def test_check_mean_only():
     check_mean_only(True)