New figure file and modified import_ccle, comments fixed.

Author: quinn-mccall

pf2barcode/figures/figure4.py

@@ -0,0 +1,80 @@
+"""
+Figure 4 -- Recreation of Boxplot of the correlation distance for related cells
+(blue), and randomly sampled cells from the GEMLI paper
+"""
+
+import pandas as pd
+import numpy as np
+import seaborn as sns
+from scipy.spatial.distance import pdist
+from pf2barcode.imports import import_CCLE
+from .common import getSetup, subplotLabel
+from sklearn.preprocessing import scale
+
+
+def makeFigure():
+    """Boxplot of correlation distance for related and random cells per lineage."""
+    ax, f = getSetup((8, 4), (1, 1))
+    subplotLabel(ax)
+
+    X = import_CCLE(pca_option="none")
+
+    # Filter out unknown or rare barcodes
+    X = X[X.obs["SW"] != "unknown"]
+    good_SW = X.obs["SW"].value_counts().index[X.obs["SW"].value_counts() > 10]
+    X = X[X.obs["SW"].isin(good_SW)]
+
+    # Convert matrix to dense for correlation computation
+    mat = X.X.toarray()
+    df = X.obs.copy()
+    df["index"] = np.arange(len(df))
+
+    results = []
+
+    for sw, cells in df.groupby("SW"):
+        idx = cells["index"].values
+        if len(idx) < 2:
+            continue
+
+        # Related (within-lineage) distances
+        related = pdist(mat[idx], metric="correlation")
+
+        # Random (same number of pairs)
+        n_pairs = len(related)
+        n_cells = mat.shape[0]
+        random_corrs = []
+        for _ in range(100):
+            pairs = np.random.choice(n_cells, (n_pairs, 2), replace=True)
+            random_corrs.extend(
+                [1 - np.corrcoef(mat[i], mat[j])[0, 1] for i, j in pairs]
+            )
+
+        results.append(
+            pd.DataFrame(
+                {
+                    "Correlation distance": np.concatenate([related, random_corrs]),
+                    "Group": ["Cell lineage"] * len(related)
+                    + ["Random cells"] * len(random_corrs),
+                    "Lineage": [sw] * (len(related) + len(random_corrs)),
+                }
+            )
+        )
+
+    df_plot = pd.concat(results, ignore_index=True)
+
+    sns.boxplot(
+        data=df_plot,
+        x="Lineage",
+        y="Correlation distance",
+        hue="Group",
+        showfliers=False,
+        palette={"Cell lineage": "#377eb8", "Random cells": "#bbbbbb"},
+        ax=ax[0],
+    )
+
+    ax[0].set_title("Correlation distance for related and random cells per lineage")
+    ax[0].set_xlabel("Lineage barcode (SW)")
+    ax[0].set_ylabel("Correlation distance")
+    ax[0].legend(title=None)
+
+    return f

pf2barcode/imports.py

@@ -135,6 +135,9 @@ def import_CCLE(pca_option="dev_pca", n_comp=10) -> AnnData:
         X = prepare_dataset_dev(X)
         X.X = scale(X.X)
         sc.pp.pca(X, n_comps=n_comp, svd_solver="arpack")
+    if pca_option == "none":
+        X = X[X.X.sum(axis=1) >= 10000]
+        sc.pp.normalize_total(X, target_sum=40000)
     else:
         X = prepare_dataset(X, geneThreshold=0.001)
         X.X = scale(X.X)