Merge pull request #186 from Intron7/make-igraph-optinal

make igraph optional to keep BDS license

Author: PauBadiaM

pyproject.toml

@@ -31,12 +31,14 @@ dependencies = [
   "xgboost",
   "seaborn",
   "requests",
-  "igraph",
   "marsilea",
   "adjustText",
   # for debug logging (referenced from the issue template)
   "session-info2",
 ]
+optional-dependencies.plot = [
+  "igraph",
+]
 optional-dependencies.dev = [
   "pre-commit",
   "twine>=4.0.2",
@@ -56,13 +58,15 @@ optional-dependencies.doc = [
   "sphinx-tabs",
   "sphinxcontrib-bibtex>=1",
   "sphinxext-opengraph",
+  "igraph",
 ]
 optional-dependencies.test = [
   "coverage",
   "pytest",
   "statsmodels",
   "gseapy",
   "scanpy",
+  "igraph",
 ]
 # https://docs.pypi.org/project_metadata/#project-urls
 urls.Documentation = "https://decoupler.readthedocs.io/"

src/decoupler/pl/_network.py

@@ -1,23 +1,51 @@
-from typing import Tuple
+from typing import TYPE_CHECKING
 
-import pandas as pd
-import numpy as np
 import matplotlib
+import matplotlib.cm
+import matplotlib.colors
+import matplotlib.gridspec
+import numpy as np
+import pandas as pd
 from matplotlib.lines import Line2D
-import matplotlib.pyplot as plt
-import igraph as ig
 
 from decoupler._docs import docs
 from decoupler._Plotter import Plotter
 
+# Handle optional igraph dependency
+try:
+    import igraph as ig
+
+    HAS_IGRAPH = True
+    if TYPE_CHECKING:
+        from igraph import Graph
+    else:
+        Graph = ig.Graph
+except ImportError:
+    ig = None
+    HAS_IGRAPH = False
+    if TYPE_CHECKING:
+        from typing import Any as Graph
+    else:
+        Graph = None
+
+
+def _check_igraph() -> None:
+    """Check if igraph is available and raise informative error if not."""
+    if not HAS_IGRAPH:
+        raise ImportError(
+            "igraph is not installed. Please install it using:\n"
+            "  pip install igraph\n"
+            "or install decoupler with plotting dependencies:\n"
+            "  pip install 'decoupler[plot]'"
+        )
+
 
 def _src_idxs(
     score: pd.DataFrame,
     sources: int | list | str,
     by_abs: bool,
 ) -> np.ndarray:
-    assert isinstance(sources, (int, list, str)), \
-    'sources must be int, list or str'
+    assert isinstance(sources, (int, list, str)), "sources must be int, list or str"
     if isinstance(sources, int):
         if by_abs:
             s_idx = np.argsort(-abs(score.values[0]))[:sources]
@@ -36,62 +64,62 @@ def _trg_idxs(
     targets: int | list | str,
     by_abs: bool,
 ) -> np.ndarray:
-    assert isinstance(targets, (int, list, str)), \
-    'targets must be int, list or str'
+    assert isinstance(targets, (int, list, str)), "targets must be int, list or str"
     if isinstance(targets, int):
-        net['prod'] = [data.iloc[0][t] * w if t in data.columns else 0 for t, w in zip(net['target'], net['weight'])]
+        net["prod"] = [
+            data.iloc[0][t] * w if t in data.columns else 0 for t, w in zip(net["target"], net["weight"], strict=False)
+        ]
         if by_abs:
-            net['prod'] = abs(net['prod'])
+            net["prod"] = abs(net["prod"])
         t_idx = (
-            net
-            .sort_values(['source', 'prod'], ascending=[True, False])
-            .groupby(['source'], observed=True)
+            net.sort_values(["source", "prod"], ascending=[True, False])
+            .groupby(["source"], observed=True)
             .head(targets)
-            .index
-            .values
+            .index.values
         )
     elif isinstance(targets, list):
-        t_idx = np.isin(net['target'].astype(str), targets)
+        t_idx = np.isin(net["target"].astype(str), targets)
     else:
-        t_idx = np.isin(net['target'].astype(str), [targets])
+        t_idx = np.isin(net["target"].astype(str), [targets])
     return t_idx
 
-    
+
 def _filter(
     data: pd.DataFrame,
     score: pd.DataFrame,
     net: pd.DataFrame,
     sources: int,
     targets: int,
     by_abs: bool,
-) -> Tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame]:
-    assert isinstance(data, pd.DataFrame), 'data must be pd.DataFrame'
-    assert isinstance(score, pd.DataFrame), 'score must be pd.DataFrame'
-    assert np.all(data.index == score.index) and (data.index.size == 1), \
-    'data and score need to have the same row index.'
-    assert isinstance(by_abs, bool), 'by_abs must be bool'
+) -> tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame]:
+    assert isinstance(data, pd.DataFrame), "data must be pd.DataFrame"
+    assert isinstance(score, pd.DataFrame), "score must be pd.DataFrame"
+    assert np.all(data.index == score.index) and (data.index.size == 1), (
+        "data and score need to have the same row index."
+    )
+    assert isinstance(by_abs, bool), "by_abs must be bool"
     # Select top sources
     s_idx = _src_idxs(score=score, sources=sources, by_abs=by_abs)
     # Filter
     score = score.iloc[:, s_idx]
-    net = net.loc[np.isin(net['source'].astype(str), score.columns.astype(str)), :].copy()
-    if 'weight' not in net.columns:
-        net['weight'] = 1.
+    net = net.loc[np.isin(net["source"].astype(str), score.columns.astype(str)), :].copy()
+    if "weight" not in net.columns:
+        net["weight"] = 1.0
     # Select top targets
     t_idx = _trg_idxs(data=data, net=net, targets=targets, by_abs=by_abs)
     # Filter
     net = net.loc[t_idx]
     # Filter unmatched features
-    data = data.loc[:, np.isin(data.columns.astype(str), net['target'].astype(str))]
-    net = net.loc[np.isin(net['target'].astype(str), data.columns.astype(str)), :]
+    data = data.loc[:, np.isin(data.columns.astype(str), net["target"].astype(str))]
+    net = net.loc[np.isin(net["target"].astype(str), data.columns.astype(str)), :]
     return data, score, net
 
 
 def _norm(
     x: np.ndarray,
     vcenter: bool,
 ) -> matplotlib.colors.Normalize:
-    assert isinstance(vcenter, bool), 'vcenter must be bool'
+    assert isinstance(vcenter, bool), "vcenter must be bool"
     if vcenter:
         vmax = np.max(np.abs(x))
         norm = matplotlib.colors.Normalize(vmin=-vmax, vmax=vmax)
@@ -105,7 +133,7 @@ def _norm(
 def _dict_types(
     data: pd.DataFrame,
     score: pd.DataFrame,
-) -> Tuple[dict, np.ndarray]:
+) -> tuple[dict, np.ndarray]:
     vs = np.unique(np.hstack([data.columns, score.columns]))
     v_dict = {k: i for i, k in enumerate(vs)}
     types = (~np.isin(vs, score.columns)) * 1
@@ -118,7 +146,7 @@ def _net_2_elist(
 ) -> list:
     edges = []
     for i in net.index:
-        source, target = net.loc[i, 'source'], net.loc[i, 'target']
+        source, target = net.loc[i, "source"], net.loc[i, "target"]
         edge = [v_dict[source], v_dict[target]]
         edges.append(edge)
     return edges
@@ -128,8 +156,9 @@ def _net_2_g(
     data: pd.DataFrame,
     score: pd.DataFrame,
     net: pd.DataFrame,
-) -> ig.Graph:
+) -> Graph:
     # Unify network
+    _check_igraph()
     v_dict, types = _dict_types(data=data, score=score)
     # Transform net to edges
     edges = _net_2_elist(net=net, v_dict=v_dict)
@@ -139,15 +168,15 @@ def _net_2_g(
         directed=True,
     )
     # Update attributes
-    g.es['weight'] = net['weight'].values
-    g.vs['type'] = types
-    g.vs['label'] = list(v_dict.keys())
-    g.vs['shape'] = np.where(types, 'circle', 'square')
+    g.es["weight"] = net["weight"].values
+    g.vs["type"] = types
+    g.vs["label"] = list(v_dict.keys())
+    g.vs["shape"] = np.where(types, "circle", "square")
     return g
 
 
 def _gcolors(
-    g: ig.Graph,
+    g: Graph,
     data: pd.DataFrame,
     score: pd.DataFrame,
     s_norm: matplotlib.colors.Normalize,
@@ -160,16 +189,16 @@ def _gcolors(
         s_cmap = matplotlib.colormaps.get_cmap(s_cmap)
         t_cmap = matplotlib.colormaps.get_cmap(t_cmap)
         color = []
-        for i, k in enumerate(g.vs['label']):
-            if g.vs['type'][i]:
+        for i, k in enumerate(g.vs["label"]):
+            if g.vs["type"][i]:
                 color.append(t_cmap(t_norm(data[k].values[0])))
             else:
                 color.append(s_cmap(s_norm(score[k].values[0])))
         is_cmap = True
     else:
-        color = [s_cmap if typ == 0. else t_cmap for typ in g.vs['type']]
+        color = [s_cmap if typ == 0.0 else t_cmap for typ in g.vs["type"]]
         is_cmap = False
-    g.vs['color'] = color
+    g.vs["color"] = color
     return is_cmap
 
 
@@ -183,15 +212,15 @@ def network(
     by_abs=True,
     size_node=5,
     size_label=2.5,
-    s_cmap='RdBu_r',
-    t_cmap='viridis',
+    s_cmap="RdBu_r",
+    t_cmap="viridis",
     vcenter=False,
-    c_pos_w='darkgreen',
-    c_neg_w='darkred',
-    s_label='Enrichment\nscore',
-    t_label='Gene\nexpression',
-    layout='kk',
-    **kwargs
+    c_pos_w="darkgreen",
+    c_neg_w="darkred",
+    s_label="Enrichment\nscore",
+    t_label="Gene\nexpression",
+    layout="kk",
+    **kwargs,
 ):
     """
     Plot results of enrichment analysis as network.
@@ -231,14 +260,16 @@ def network(
         Layout to use to order the nodes. Check ``igraph`` documentation for more options.
     %(plot)s
     """
-    assert isinstance(net, pd.DataFrame), 'net must be pd.DataFrame'
-    assert (data is None) == (score is None), 'data and score must either both be None'
+    assert isinstance(net, pd.DataFrame), "net must be pd.DataFrame"
+    assert (data is None) == (score is None), "data and score must either both be None"
+    if ig is None:
+        raise ImportError("igraph is not installed. Please install it using `pip install igraph`.")
     if data is None:
-        srcs = net['source'].unique().astype('U')
-        score = pd.DataFrame(np.ones((1, srcs.size)), index=['0'], columns=srcs)
-        trgs = net['target'].unique().astype('U')
-        data = pd.DataFrame(np.ones((1, trgs.size)), index=['0'], columns=trgs)
-        t_cmap = 'white'
+        srcs = net["source"].unique().astype("U")
+        score = pd.DataFrame(np.ones((1, srcs.size)), index=["0"], columns=srcs)
+        trgs = net["target"].unique().astype("U")
+        data = pd.DataFrame(np.ones((1, trgs.size)), index=["0"], columns=trgs)
+        t_cmap = "white"
     # Filter
     fdata, fscore, fnet = _filter(
         data=data,
@@ -253,7 +284,7 @@ def network(
     t_norm = _norm(x=fdata, vcenter=vcenter)
     # Make graph
     g = _net_2_g(data=fdata, score=fscore, net=fnet)
-    g.es['color'] = [c_pos_w if w > 0 else c_neg_w for w in g.es['weight']]
+    g.es["color"] = [c_pos_w if w > 0 else c_neg_w for w in g.es["weight"]]
     is_cmap = _gcolors(
         g=g,
         data=data,
@@ -264,7 +295,7 @@ def network(
         t_cmap=t_cmap,
     )
     # Instance
-    kwargs['ax'] = None
+    kwargs["ax"] = None
     bp = Plotter(**kwargs)
     bp.fig.delaxes(bp.ax)
     # Plot
@@ -279,7 +310,7 @@ def network(
         layout=layout,
         vertex_size=(size_node * bp.dpi) / (bp.figsize[0] * bp.figsize[0]),
         vertex_size_label=(size_label * bp.dpi) / (bp.figsize[0] * bp.figsize[0]),
-        bbox_inches='tight',
+        bbox_inches="tight",
     )
     if is_cmap:
         sm = matplotlib.cm.ScalarMappable(norm=s_norm, cmap=s_cmap)
@@ -290,19 +321,35 @@ def network(
         ax2.axis("off")
         ax4.axis("off")
     # Add legend
-    square = Line2D([0], [0], marker='s', color='w', label='Source', markerfacecolor='white',
-                    markeredgecolor='black', markersize=10)
-    circle = Line2D([0], [0], marker='o', color='w', label='Target', markerfacecolor='white',
-                    markeredgecolor='black', markersize=10)
-    line1 = Line2D((0, 0), (1, 0), color=c_pos_w, lw=2, marker='>',)
-    line2 = Line2D((0, 0), (1, 0), color=c_neg_w, lw=2, marker='>',)
+    square = Line2D(
+        [0], [0], marker="s", color="w", label="Source", markerfacecolor="white", markeredgecolor="black", markersize=10
+    )
+    circle = Line2D(
+        [0], [0], marker="o", color="w", label="Target", markerfacecolor="white", markeredgecolor="black", markersize=10
+    )
+    line1 = Line2D(
+        (0, 0),
+        (1, 0),
+        color=c_pos_w,
+        lw=2,
+        marker=">",
+    )
+    line2 = Line2D(
+        (0, 0),
+        (1, 0),
+        color=c_neg_w,
+        lw=2,
+        marker=">",
+    )
     handles = [square, circle, line1, line2]
-    labels = ['Source', 'Target', 'Positive', 'Negative']
+    labels = ["Source", "Target", "Positive", "Negative"]
     legend = ax3.legend(
         handles=[square, circle, line1, line2],
         labels=labels,
         frameon=False,
-        loc='center', bbox_to_anchor=(0.5, 0.5), bbox_transform=ax3.transAxes
+        loc="center",
+        bbox_to_anchor=(0.5, 0.5),
+        bbox_transform=ax3.transAxes,
     )
-    ax3.axis('off')
+    ax3.axis("off")
     return bp._return()