""" plot_util.py — Plotting utilities for Squidiff reproducibility notebooks. The original authors used a local private script that was not included in the public repository. This file reimplements the required functions based on their call signatures in the reproducibility notebooks. Functions: plot_pca -- PCA scatter plot of embeddings colored by group label. display_reconst -- Scatter + kernel density comparison of original vs. reconstructed gene expressions. """ import numpy as np import matplotlib.pyplot as plt import matplotlib.patches as mpatches from sklearn.decomposition import PCA import torch def plot_pca( data, label=None, size=3, alpha=0.8, colorlist=None, color_label=None, title=None, figsize=(3, 3), dpi=300, ): """ Project data to 2D via PCA and render a scatter plot colored by label. Parameters ---------- data : torch.Tensor or numpy.ndarray Input matrix of shape (n_cells, n_features). label : array-like, optional Group labels for each cell (e.g., obs['Group']). Used for coloring. size : float, optional Marker size for scatter points. Default is 3. alpha : float, optional Marker opacity. Default is 0.8. colorlist : list of str, optional List of hex color strings, one per unique label value. Default cycles through a built-in palette. color_label : list, optional Subset of label values to include in the legend. If None, all unique label values are shown. title : str, optional Plot title. If None, no title is shown. figsize : tuple, optional Figure size in inches. Default is (3, 3). dpi : int, optional Figure DPI. Default is 300. """ # Convert torch.Tensor to numpy if necessary if isinstance(data, torch.Tensor): data = data.detach().cpu().numpy() else: data = np.array(data) # Reduce to 2D via PCA if data.shape[1] > 2: pca = PCA(n_components=2) coords = pca.fit_transform(data) else: coords = data # Already 2D # Resolve labels if label is not None: labels = np.array(label) unique_labels = np.unique(labels) else: labels = np.zeros(len(coords), dtype=int) unique_labels = np.array([0]) # Resolve colors default_palette = [ '#3145a8', '#fa2616', '#40a8f7', '#f5bf36', '#2ca02c', '#9467bd', '#8c564b', '#e377c2', ] if colorlist is None: colorlist = default_palette # Determine which labels to show in legend if color_label is not None: legend_labels = [str(v) for v in color_label] else: legend_labels = [str(v) for v in unique_labels] # Plot fig, ax = plt.subplots(figsize=figsize, dpi=dpi) handles = [] for idx, ul in enumerate(unique_labels): mask = labels == ul color = colorlist[idx % len(colorlist)] ax.scatter( coords[mask, 0], coords[mask, 1], c=color, s=size, alpha=alpha, rasterized=True, ) if str(ul) in legend_labels: handles.append( mpatches.Patch(color=color, label=str(ul)) ) ax.set_xlabel('PC1') ax.set_ylabel('PC2') if title: ax.set_title(title) if handles: ax.legend(handles=handles, markerscale=2, frameon=False, bbox_to_anchor=(1.01, 1), loc='upper left') plt.tight_layout() plt.show() def display_reconst( original_df, reconstructed_df, density=False, size=(2.5, 2.5), dpi=300, alpha=0.3, color_orig='#3145a8', color_reconst='#fa2616', xlabel='Original gene expression', ylabel='Reconstructed gene expression', ): """ Display a scatter plot comparing original and reconstructed gene expression, with optional kernel-density overlays. Parameters ---------- original_df : pandas.DataFrame or array-like Flattened original gene expression values. reconstructed_df : pandas.DataFrame or array-like Flattened reconstructed gene expression values. density : bool, optional If True, overlay a KDE (kernel density estimate) on 1-D marginals. Default is False. size : tuple, optional Figure size (width, height) in inches. Default is (2.5, 2.5). dpi : int, optional Figure DPI. Default is 300. alpha : float, optional Scatter point opacity. Default is 0.3. color_orig : str, optional Color for original values in density plot. Default is '#3145a8'. color_reconst : str, optional Color for reconstructed values in density plot. Default is '#fa2616'. xlabel : str, optional X-axis label. Default is 'Original gene expression'. ylabel : str, optional Y-axis label. Default is 'Reconstructed gene expression'. """ import pandas as pd from scipy.stats import gaussian_kde # Flatten to 1-D numpy arrays orig = np.array(original_df).flatten() reconst = np.array(reconstructed_df).flatten() if density: # Two-panel layout: scatter (left) + KDE (right) fig, axes = plt.subplots(1, 2, figsize=(size[0] * 2, size[1]), dpi=dpi) # Left: scatter axes[0].scatter(reconst, orig, s=2, alpha=alpha, color=color_orig, rasterized=True) axes[0].set_xlabel('Reconstructed') axes[0].set_ylabel('Original') # Diagonal reference line lims = [min(orig.min(), reconst.min()), max(orig.max(), reconst.max())] axes[0].plot(lims, lims, 'k--', linewidth=0.8, alpha=0.5) # Right: KDE of original vs reconstructed for vals, color, lbl in [ (orig, color_orig, 'Original'), (reconst, color_reconst, 'Reconstructed'), ]: kde = gaussian_kde(vals, bw_method=0.3) x_range = np.linspace(vals.min(), vals.max(), 300) axes[1].plot(x_range, kde(x_range), color=color, label=lbl) axes[1].fill_between(x_range, kde(x_range), alpha=0.2, color=color) axes[1].set_xlabel('Gene expression') axes[1].set_ylabel('Density') axes[1].legend(frameon=False, fontsize=8) else: # Single scatter plot fig, ax = plt.subplots(figsize=size, dpi=dpi) ax.scatter(reconst, orig, s=2, alpha=alpha, color=color_orig, rasterized=True) lims = [min(orig.min(), reconst.min()), max(orig.max(), reconst.max())] ax.plot(lims, lims, 'k--', linewidth=0.8, alpha=0.5) ax.set_xlabel(xlabel) ax.set_ylabel(ylabel) plt.tight_layout() plt.show()