Heatmap edits

cellcommunicationpf2/figures/figureA4g.py

@@ -8,6 +8,7 @@
     getSetup,
 )
 import numpy as np
+import pandas as pd
 import seaborn as sns
 from ..utils import (
     expression_product_matrix,
@@ -22,66 +23,88 @@ def makeFigure():
     ccc_rise_cmp = 5
 
     X_mdc_sender = X[(X.obs["broad_cell_type"] == "CD4 T cells")]
+    sender_weights = X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1]
     X_mdc_sender = X_mdc_sender[
-        np.argsort(-X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1])
+        np.argsort(-np.abs(sender_weights))
     ]
 
     X_mdc_receiver = X[(X.obs["broad_cell_type"] == "Dendritic Cells")]
-
+    receiver_weights = X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1]
     X_mdc_receiver = X_mdc_receiver[
-        np.argsort(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1])
+        np.argsort(np.abs(receiver_weights))
     ]
 
 
     pairs = ["CD40LG", "CD40"]
     df = expression_product_matrix(X_mdc_sender, X_mdc_receiver, pairs[0], pairs[1])
     df_grouped = group_matrix(df)
-    # Keep max value consistent across heatmaps for better comparison
     sns.heatmap(df_grouped, ax=ax[0], cmap="rocket")
     ax[0].set_title(f"{pairs[0]}-{pairs[1]} Interaction")
-    ax[0].set_xlabel("Receiver Epithelial Cells")
+    ax[0].set_xlabel("Receiver DCs")
     ax[0].set_ylabel("Sender CD4 T Cells")
-
-
-
+
+    # Weight verification: sender + receiver component weight sums
+    s_sorted = np.abs(sender_weights[np.argsort(-np.abs(sender_weights))])
+    r_sorted = np.abs(receiver_weights[np.argsort(np.abs(receiver_weights))])
+    weight_matrix = group_weight_sum(s_sorted, r_sorted)
+    sns.heatmap(weight_matrix, ax=ax[2], cmap="viridis")
+    ax[2].set_title("Component Weight (sender + receiver)")
+    ax[2].set_xlabel("Receiver DCs")
+    ax[2].set_ylabel("Sender CD4 T Cells")
+
+
     X_mdc_sender = X[(X.obs["broad_cell_type"] == "Epithelial cells")]
     X_mdc_sender = X_mdc_sender[
-        np.argsort(-X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1])
+        np.argsort(-np.abs(X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1]))
     ]
 
     X_mdc_receiver = X[(X.obs["broad_cell_type"] == "Epithelial cells")]
 
     X_mdc_receiver = X_mdc_receiver[
-        np.argsort(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1])
+        np.argsort(np.abs(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1]))
     ]
 
     df = expression_product_matrix(X_mdc_sender, X_mdc_receiver, pairs[0], pairs[1])
     df_grouped = group_matrix(df)
     sns.heatmap(df_grouped, ax=ax[1], cmap="rocket")
     ax[1].set_title(f"{pairs[0]}-{pairs[1]} Interaction")
     ax[1].set_xlabel("Receiver Epithelial Cells")
-    ax[1].set_ylabel("Sender Macrophages")
+    ax[1].set_ylabel("Sender Epithelial Cells")
+
+    # Component weight distributions for panel (a) cell types
+    ax[3].hist(sender_weights, bins=30, alpha=0.6, label=f"Sender sc_B (CD4 T)\nn={len(sender_weights)}, {np.sum(sender_weights<0)} neg")
+    ax[3].hist(receiver_weights, bins=30, alpha=0.6, label=f"Receiver rc_C (DCs)\nn={len(receiver_weights)}, {np.sum(receiver_weights<0)} neg")
+    ax[3].axvline(0, color="black", linestyle="--", linewidth=0.5)
+    ax[3].set_xlabel("Component Weight")
+    ax[3].set_ylabel("Count")
+    ax[3].set_title(f"Cmp {ccc_rise_cmp} Weight Distribution")
+    ax[3].legend(fontsize=6)
 
     return f
 
 
 def group_matrix(df):
     """
     Groups a DataFrame into a 10x10 matrix by binning rows and columns and averaging within bins.
-    Prints shape information for debugging.
     """
-    print(f"Original matrix shape: {df.shape}")
     n_rows = len(df)
     n_cols = len(df.columns)
     row_group_size = n_rows // 10
     col_group_size = n_cols // 10
-    print(f"Row group size: {row_group_size}, Col group size: {col_group_size}")
     row_groups = np.arange(n_rows) // row_group_size
     col_groups = np.arange(n_cols) // col_group_size
     row_groups = np.clip(row_groups, 0, 9)
     col_groups = np.clip(col_groups, 0, 9)
     df_grouped = df.groupby(row_groups).mean()
     df_grouped = df_grouped.groupby(col_groups, axis=1).mean()
-    print(f"Final grouped matrix shape: {df_grouped.shape}")
-    print(df_grouped)
-    return df_grouped
+    return df_grouped
+
+
+def group_weight_sum(sender_weights, receiver_weights):
+    """Bin weights into 10 groups each, return 10x10 matrix of sender + receiver means."""
+    n_s, n_r = len(sender_weights), len(receiver_weights)
+    s_groups = np.clip(np.arange(n_s) // (n_s // 10), 0, 9)
+    r_groups = np.clip(np.arange(n_r) // (n_r // 10), 0, 9)
+    s_means = pd.Series(sender_weights).groupby(s_groups).mean().values
+    r_means = pd.Series(receiver_weights).groupby(r_groups).mean().values
+    return pd.DataFrame(s_means[:, None] + r_means[None, :])

cellcommunicationpf2/figures/figureA4h.py

@@ -0,0 +1,141 @@
+"""
+Figure A4h: Heatmaps of ligand-receptor expression products for LTA&LTB-LTBR
+"""
+
+import anndata
+from .common import (
+    subplotLabel,
+    getSetup,
+)
+import numpy as np
+import pandas as pd
+import seaborn as sns
+from ..utils import (
+    expression_product_matrix,
+)
+
+
+def makeFigure():
+    ax, f = getSetup((6, 6), (2, 2))
+    subplotLabel(ax)
+
+    X = anndata.read_h5ad("/opt/andrew/ccc/bal_alad.h5ad")
+    ccc_rise_cmp = 5
+
+    X_mdc_sender = X[(X.obs["broad_cell_type"] == "CD4 T cells")]
+    sender_weights = X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1]
+    X_mdc_sender = X_mdc_sender[
+        np.argsort(-np.abs(sender_weights))
+    ]
+
+    X_mdc_receiver = X[(X.obs["broad_cell_type"] == "Dendritic Cells")]
+    receiver_weights = X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1]
+    X_mdc_receiver = X_mdc_receiver[
+        np.argsort(np.abs(receiver_weights))
+    ]
+
+
+    # LTA&LTB-LTBR interaction: ligand = min(LTA, LTB) per cell (complex aggregation)
+    df = expression_product_matrix_complex(
+        X_mdc_sender, X_mdc_receiver, ["LTA", "LTB"], "LTBR"
+    )
+    df_grouped = group_matrix(df)
+    sns.heatmap(df_grouped, ax=ax[0], cmap="rocket")
+    ax[0].set_title("LTA&LTB-LTBR Interaction")
+    ax[0].set_xlabel("Receiver DCs")
+    ax[0].set_ylabel("Sender CD4 T Cells")
+
+    # Weight verification: sender + receiver component weight sums
+    s_sorted = np.abs(sender_weights[np.argsort(-np.abs(sender_weights))])
+    r_sorted = np.abs(receiver_weights[np.argsort(np.abs(receiver_weights))])
+    weight_matrix = group_weight_sum(s_sorted, r_sorted)
+    sns.heatmap(weight_matrix, ax=ax[2], cmap="viridis")
+    ax[2].set_title("Component Weight (sender + receiver)")
+    ax[2].set_xlabel("Receiver DCs")
+    ax[2].set_ylabel("Sender CD4 T Cells")
+
+
+    X_mdc_sender = X[(X.obs["broad_cell_type"] == "Epithelial cells")]
+    X_mdc_sender = X_mdc_sender[
+        np.argsort(-np.abs(X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1]))
+    ]
+
+    X_mdc_receiver = X[(X.obs["broad_cell_type"] == "Epithelial cells")]
+
+    X_mdc_receiver = X_mdc_receiver[
+        np.argsort(np.abs(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1]))
+    ]
+
+    df = expression_product_matrix_complex(
+        X_mdc_sender, X_mdc_receiver, ["LTA", "LTB"], "LTBR"
+    )
+    df_grouped = group_matrix(df)
+    sns.heatmap(df_grouped, ax=ax[1], cmap="rocket")
+    ax[1].set_title("LTA&LTB-LTBR Interaction")
+    ax[1].set_xlabel("Receiver Epithelial Cells")
+    ax[1].set_ylabel("Sender Epithelial Cells")
+
+    # Component weight distributions
+    ax[3].hist(sender_weights, bins=30, alpha=0.6, label=f"Sender sc_B (CD4 T)\nn={len(sender_weights)}, {np.sum(sender_weights<0)} neg")
+    ax[3].hist(receiver_weights, bins=30, alpha=0.6, label=f"Receiver rc_C (DCs)\nn={len(receiver_weights)}, {np.sum(receiver_weights<0)} neg")
+    ax[3].axvline(0, color="black", linestyle="--", linewidth=0.5)
+    ax[3].set_xlabel("Component Weight")
+    ax[3].set_ylabel("Count")
+    ax[3].set_title(f"Cmp {ccc_rise_cmp} Weight Distribution")
+    ax[3].legend(fontsize=6)
+
+    return f
+
+
+def expression_product_matrix_complex(X1, X2, ligand_genes, receptor):
+    """
+    Like expression_product_matrix but handles complex ligands.
+    For a complex (e.g. LTA&LTB), ligand expression = min(gene1, gene2, ...) per cell.
+    """
+    # Get ligand expression as min across complex subunits
+    lig_exprs = []
+    for gene in ligand_genes:
+        expr = X1[:, gene].X
+        if hasattr(expr, "toarray"):
+            expr = expr.toarray().flatten()
+        else:
+            expr = np.ravel(np.array(expr))
+        lig_exprs.append(expr)
+    ligand_expr = np.minimum.reduce(lig_exprs)
+
+    # Get receptor expression
+    rec_expr = X2[:, receptor].X
+    if hasattr(rec_expr, "toarray"):
+        rec_expr = rec_expr.toarray().flatten()
+    else:
+        rec_expr = np.ravel(np.array(rec_expr))
+
+    product_matrix = np.outer(ligand_expr, rec_expr)
+    return pd.DataFrame(product_matrix, index=X1.obs_names, columns=X2.obs_names)
+
+
+def group_matrix(df):
+    """
+    Groups a DataFrame into a 10x10 matrix by binning rows and columns and averaging within bins.
+    """
+    n_rows = len(df)
+    n_cols = len(df.columns)
+    row_group_size = n_rows // 10
+    col_group_size = n_cols // 10
+    row_groups = np.arange(n_rows) // row_group_size
+    col_groups = np.arange(n_cols) // col_group_size
+    row_groups = np.clip(row_groups, 0, 9)
+    col_groups = np.clip(col_groups, 0, 9)
+    df_grouped = df.groupby(row_groups).mean()
+    df_grouped = df_grouped.groupby(col_groups, axis=1).mean()
+    return df_grouped
+
+
+def group_weight_sum(sender_weights, receiver_weights):
+    """Bin weights into 10 groups each, return 10x10 matrix of sender + receiver means."""
+    n_s, n_r = len(sender_weights), len(receiver_weights)
+    s_groups = np.clip(np.arange(n_s) // (n_s // 10), 0, 9)
+    r_groups = np.clip(np.arange(n_r) // (n_r // 10), 0, 9)
+    s_means = pd.Series(sender_weights).groupby(s_groups).mean().values
+    r_means = pd.Series(receiver_weights).groupby(r_groups).mean().values
+    return pd.DataFrame(s_means[:, None] + r_means[None, :])

cellcommunicationpf2/figures/figureS8a_b.py

@@ -8,10 +8,9 @@
     getSetup,
 )
 import numpy as np
+import pandas as pd
 import seaborn as sns
 from ..utils import (
-    add_obs_cmp_label,
-    add_obs_cmp_unique_one,
     expression_product_matrix,
 )
 
@@ -24,14 +23,15 @@ def makeFigure():
     ccc_rise_cmp = 13
 
     X_mdc_sender = X[X.obs["broad_cell_type"] == "Dendritic Cells"]
+    sender_weights = X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1]
     X_mdc_sender = X_mdc_sender[
-        np.argsort(-X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1])
+        np.argsort(-np.abs(sender_weights))
     ]
 
     X_mdc_receiver = X[(X.obs["broad_cell_type"] == "Dendritic Cells")]
-
+    receiver_weights = X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1]
     X_mdc_receiver = X_mdc_receiver[
-        np.argsort(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1])
+        np.argsort(np.abs(receiver_weights))
     ]
 
 
@@ -70,18 +70,27 @@ def makeFigure():
         ax[i].set_xlabel("Dendritic Cells")
         ax[i].set_ylabel("Dendritic Cells")
 
+    # Weight verification: sender + receiver component weight sums
+    s_sorted = np.abs(sender_weights[np.argsort(-np.abs(sender_weights))])
+    r_sorted = np.abs(receiver_weights[np.argsort(np.abs(receiver_weights))])
+    weight_matrix = group_weight_sum(s_sorted, r_sorted)
+    sns.heatmap(weight_matrix, ax=ax[1], cmap="viridis")
+    ax[1].set_title("Component Weight (sender + receiver)")
+    ax[1].set_xlabel("Dendritic Cells")
+    ax[1].set_ylabel("Dendritic Cells")
+
 
     ccc_rise_cmp = 13
 
     X_mdc_sender = X[X.obs["broad_cell_type"] == "Dendritic Cells"]
     X_mdc_sender = X_mdc_sender[
-        np.argsort(-X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1])
+        np.argsort(-np.abs(X_mdc_sender.obsm["sc_B"][:, ccc_rise_cmp - 1]))
     ]
 
     X_mdc_receiver = X[(X.obs["broad_cell_type"] == "Epithelial cells")]
 
     X_mdc_receiver = X_mdc_receiver[
-        np.argsort(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1])
+        np.argsort(np.abs(X_mdc_receiver.obsm["rc_C"][:, ccc_rise_cmp - 1]))
     ]
 
 
@@ -120,7 +129,23 @@ def makeFigure():
         ax[i+2].set_xlabel("Epithelial Cells")
         ax[i+2].set_ylabel("Dendritic Cells")
 
-
-
-
-    return f
+    # Component weight distributions
+    ax[3].hist(sender_weights, bins=30, alpha=0.6, label=f"Sender sc_B (DCs)\nn={len(sender_weights)}, {np.sum(sender_weights<0)} neg")
+    ax[3].hist(receiver_weights, bins=30, alpha=0.6, label=f"Receiver rc_C (DCs)\nn={len(receiver_weights)}, {np.sum(receiver_weights<0)} neg")
+    ax[3].axvline(0, color="black", linestyle="--", linewidth=0.5)
+    ax[3].set_xlabel("Component Weight")
+    ax[3].set_ylabel("Count")
+    ax[3].set_title(f"Cmp {ccc_rise_cmp} Weight Distribution")
+    ax[3].legend(fontsize=6)
+
+    return f
+
+
+def group_weight_sum(sender_weights, receiver_weights):
+    """Bin weights into 10 groups each, return 10x10 matrix of sender + receiver means."""
+    n_s, n_r = len(sender_weights), len(receiver_weights)
+    s_groups = np.clip(np.arange(n_s) // (n_s // 10), 0, 9)
+    r_groups = np.clip(np.arange(n_r) // (n_r // 10), 0, 9)
+    s_means = pd.Series(sender_weights).groupby(s_groups).mean().values
+    r_means = pd.Series(receiver_weights).groupby(r_groups).mean().values
+    return pd.DataFrame(s_means[:, None] + r_means[None, :])