Merge pull request #300 from lilab-bcb/pseudo-deseq2

Reorganize pseudobulk and deseq2 functions

pegasus/plotting/plot_library.py

@@ -353,7 +353,7 @@ def scatter(
                                 fontsize=legend_fontsize[attr_id],
                                 ncol=_get_legend_ncol(label_size, legend_ncol),
                             )
-                            for handle in legend.legendHandles:
+                            for handle in legend.legend_handles:
                                 handle.set_sizes([300.0 if scale_factor is None else 100.0])
                         elif legend_loc[attr_id] == "on data":
                             texts = []
@@ -588,7 +588,7 @@ def scatter_groups(
                             fontsize=legend_fontsize,
                             ncol=legend_ncol,
                         )
-                        for handle in legend.legendHandles:
+                        for handle in legend.legend_handles:
                             handle.set_sizes([300.0])
                     elif legend_loc == "on data":
                         texts = []
@@ -2092,7 +2092,7 @@ def volcano(
         fontsize=8,
         ncol=4,
     )
-    for handle in legend.legendHandles: # adjust legend size
+    for handle in legend.legend_handles: # adjust legend size
         handle.set_sizes([50.0])
 
     ax.axhline(y = yconst, c = 'k', lw = 0.5, ls = '--')
@@ -2355,7 +2355,7 @@ def wordcloud(
     >>> fig = pg.wordcloud(data, factor=0)
     """
     fig, ax = _get_subplot_layouts(panel_size=panel_size, dpi=dpi) # default nrows = 1 & ncols = 1
-    
+
     assert 'W' in data.uns
     hvg = data.var_names[data.var[features]]
     word_dict = {}

pegasus/pseudo/convenient.py

@@ -49,7 +49,7 @@ def markers(
     if de_key not in pseudobulk.varm.keys():
         raise ValueError("Please run DE analysis first")
 
-    res_dict = {}    
+    res_dict = {}
     df = pd.DataFrame(data=pseudobulk.varm[de_key], index=pseudobulk.var_names)
     idx = df["padj"] <= alpha
 
@@ -206,11 +206,20 @@ def volcano(
         logger.warning("Please conduct DE test first!")
         return None
 
-    log2fc = de_res[fcstr]
-    pvals = de_res[pstr]
+    qvals = de_res[qstr]
+    # Ignore genes with q-value NaN
+    idx_select = np.where(~np.isnan(qvals))[0]
+    if idx_select.size == 0:
+        logger.warning("All genes have NaN adjusted p-values!")
+        return None
+    qvals = qvals[idx_select]
+    log2fc = de_res[fcstr][idx_select]
+    pvals = de_res[pstr][idx_select]
+    gene_names = pseudobulk.var_names[idx_select]
+
     pvals[pvals == 0.0] = 1e-45 # very small pvalue to avoid log10 0
     neglog10p = -np.log10(pvals)
-    yconst = min(neglog10p[de_res[qstr] <= qval_threshold])
+    yconst = min(neglog10p[qvals <= qval_threshold])
 
     from pegasus.plotting.plot_utils import _get_subplot_layouts
     fig, ax = _get_subplot_layouts(panel_size=panel_size, dpi=dpi)
@@ -239,7 +248,7 @@ def volcano(
         fontsize=8,
         ncol=4,
     )
-    for handle in legend.legendHandles: # adjust legend size
+    for handle in legend.legend_handles: # adjust legend size
         handle.set_sizes([50.0])
 
     ax.axhline(y = yconst, c = 'k', lw = 0.5, ls = '--')
@@ -252,13 +261,13 @@ def volcano(
     posvec = np.argsort(log2fc[idx])[::-1][0:top_n]
     for pos in posvec:
         gid = idx[pos]
-        texts.append(ax.text(log2fc[gid], neglog10p[gid], pseudobulk.var_names[gid], fontsize=5))
+        texts.append(ax.text(log2fc[gid], neglog10p[gid], gene_names[gid], fontsize=5))
 
     idx = np.where(idxsig & (log2fc <= -log2fc_threshold))[0]
     posvec = np.argsort(log2fc[idx])[0:top_n]
     for pos in posvec:
         gid = idx[pos]
-        texts.append(ax.text(log2fc[gid], neglog10p[gid], pseudobulk.var_names[gid], fontsize=5))
+        texts.append(ax.text(log2fc[gid], neglog10p[gid], gene_names[gid], fontsize=5))
 
     from adjustText import adjust_text
     adjust_text(texts, arrowprops=dict(arrowstyle='-', color='k', lw=0.5))

pegasus/tools/clustering.py

@@ -11,7 +11,7 @@
 from sklearn.cluster import KMeans
 from typing import List, Optional, Union
 
-from pegasus.tools import eff_n_jobs, construct_graph, calc_stat_per_batch, X_from_rep, slicing
+from pegasus.tools import eff_n_jobs, construct_graph, calc_stat_per_batch, update_rep, X_from_rep, slicing
 
 import logging
 logger = logging.getLogger(__name__)
@@ -780,10 +780,8 @@ def calc_dendrogram(
     """
     # Set up embedding or count matrix to use
     if genes is None:
-        if rep:
-            embed_df = pd.DataFrame(X_from_rep(data, rep))
-        else:
-            embed_df = pd.DataFrame(data.X.toarray() if issparse(data.X) else data.X)
+        rep = update_rep(rep)
+        embed_df = pd.DataFrame(X_from_rep(data, rep))
     else:
         embed_df = pd.DataFrame(slicing(data[:, genes].X))
 
@@ -825,4 +823,4 @@ def calc_dendrogram(
     np.fill_diagonal(dissim_df.to_numpy(), 0)    # Enforce main diagonal to be 0 to pass squareform requirement
     Z = linkage(squareform(dissim_df), method=linkage_method, optimal_ordering=True)
 
-    data.uns[res_key] = (Z, csi_df)    
+    data.uns[res_key] = (Z, csi_df)