From 873197a4b7a9cf2507c47d1c2dbb2896d5e359aa Mon Sep 17 00:00:00 2001 From: Pablo Moreno Date: Mon, 14 Oct 2024 13:17:26 +0100 Subject: [PATCH] Improvements to qc script --- .../scanpy/scripts/sc_qc_metrics.py | 252 ++++++++++++++++-- 1 file changed, 231 insertions(+), 21 deletions(-) diff --git a/tools/tertiary-analysis/scanpy/scripts/sc_qc_metrics.py b/tools/tertiary-analysis/scanpy/scripts/sc_qc_metrics.py index 6834b463..ab395c0f 100644 --- a/tools/tertiary-analysis/scanpy/scripts/sc_qc_metrics.py +++ b/tools/tertiary-analysis/scanpy/scripts/sc_qc_metrics.py @@ -2,7 +2,8 @@ import matplotlib.pyplot as plt import scanpy as sc -import seaborn as sns + +# import seaborn as sns def main(): @@ -12,7 +13,10 @@ def main(): ) parser.add_argument("adata_file", type=str, help="Path to AnnData object file") parser.add_argument( - "sample_field", type=str, help="Field in the obs for the sample identifier" + "--sample_field", + type=str, + default="Sample_ID", + help="Field in the obs for the sample identifier" ) parser.add_argument( "--output_format", @@ -28,6 +32,34 @@ def main(): metavar=("width", "height"), help="Size of the plots", ) + # add an argument for general plot title font size + parser.add_argument( + "--title_font_size", + type=int, + default=12, + help="General plot title font size", + ) + # add an argument for general plot label font size + parser.add_argument( + "--label_font_size", + type=int, + default=8, + help="General plot label font size", + ) + # add an argument for general plot legend font size + parser.add_argument( + "--legend_font_size", + type=int, + default=10, + help="General plot legend font size", + ) + # add an argument for the gene symbols field + parser.add_argument( + "--gene_symbols_field", + type=str, + default="gene_symbols", + help="Field in the var for the gene symbols", + ) parser.add_argument( "--percent_mito_field", type=str, @@ -58,6 +90,13 @@ def main(): default="doublet_score", help="Field in the obs for the doublet score", ) + # add an argument for an embedding to plot the cells + parser.add_argument( + "--embedding", + type=str, + default=None, + help="Embedding to plot the cells", + ) args = parser.parse_args() # Load AnnData object @@ -67,8 +106,14 @@ def main(): if args.plot_size: sc.settings.figsize = tuple(args.plot_size) - # Set output format - sc.settings.set_figure_params(format=args.output_format) + # set scanpy general plot font size and output format + sc.settings.set_figure_params(scanpy=True, + fontsize=args.label_font_size, + format=args.output_format) + # disable FutureWarning + import warnings + + warnings.simplefilter(action="ignore", category=FutureWarning) run_quality_control = False if "n_genes_by_counts" not in adata.obs.columns: @@ -77,13 +122,30 @@ def main(): run_quality_control = True qc_vars = [] + fields = [ + "n_genes_by_counts", + "total_counts", + args.percent_mito_field, + args.percent_ribo_field + ] # calculate mitochondrial genes if not provided if args.percent_mito_field not in adata.obs.columns: qc_vars.append(args.mito_field) # calculate ribo metrics if not provided + if args.ribo_field not in adata.var.columns: + # create a new column with the name args.ribo_field where genes that + # have in the gene symbols field the pattern ^RP[SL] are + # marked as true + print(f"Creating {args.ribo_field} column") + adata.var[args.ribo_field] = adata.var[args.gene_symbols_field].str.contains( + "^RP[SL]" + ) + print(f"Number of ribosomal genes: {adata.var[args.ribo_field].sum()}") if args.percent_ribo_field not in adata.obs.columns: qc_vars.append(args.ribo_field) + print(f"Calculating QC metrics for {len(qc_vars)} variables") + if len(qc_vars) > 0 or run_quality_control: sc.pp.calculate_qc_metrics( adata, @@ -95,16 +157,54 @@ def main(): adata.obs["n_genes"] = adata.obs["n_genes_by_counts"] adata.var["n_counts"] = adata.var["total_counts"] adata.var["n_cells"] = adata.var["n_cells_by_counts"] + + # Define thresholds + high_umi_threshold = adata.obs['n_counts'].quantile(0.95) # Top 5% most UMI counts + low_umi_threshold = adata.obs['n_counts'].quantile(0.05) # Bottom 5% least UMI counts + high_mito_threshold = adata.obs[args.percent_mito_field].quantile(0.90) # Top 10% pct mitochondrial genes + + from sklearn.linear_model import LinearRegression + from sklearn.preprocessing import PolynomialFeatures + + # Polynomial regression to account for curvature in the n_counts vs. n_genes relationship + poly = PolynomialFeatures(degree=2) + X_poly = poly.fit_transform(adata.obs[['n_counts']]) + model = LinearRegression() + model.fit(X_poly, adata.obs['n_genes']) + predicted_counts = model.predict(X_poly) + + # Calculate residuals + residuals = adata.obs['n_genes'] - predicted_counts + outlier_threshold = residuals.abs().quantile(0.95) # Top 5% residuals as outliers + + # Initialize diagnosis column + adata.obs['auto_diagnosis'] = 'Healthy' + + # Identify outliers + outliers = residuals.abs() > outlier_threshold + adata.obs.loc[outliers, 'auto_diagnosis'] = 'Outlier' + + + # Identify stressed/dying/apoptotic cells + stressed_cells = (adata.obs['n_counts'] > high_umi_threshold) & (adata.obs[args.percent_mito_field] > high_mito_threshold) + adata.obs.loc[stressed_cells, 'auto_diagnosis'] = 'Stressed/Dying/Apoptotic' + + # Identify poor-quality cells + poor_quality_cells = (adata.obs['n_counts'] < low_umi_threshold) & (adata.obs[args.percent_mito_field] > high_mito_threshold) + adata.obs.loc[poor_quality_cells, 'auto_diagnosis'] = 'Poor-Quality' + + # Print diagnosis summary + print(adata.obs['auto_diagnosis'].value_counts()) + # make a barplot of the auto_diagnosis, omitting the healthy cells from the plot + # but writing the number of healthy cells in the title. Plot per sample + healthy_cells = adata.obs['auto_diagnosis'] == 'Healthy' + healthy_count = healthy_cells.sum() + # General quality for whole dataset plt.figure() ax = sc.pl.violin( adata, - [ - "n_genes_by_counts", - "total_counts", - args.percent_mito_field, - args.percent_ribo_field, - ], + fields, jitter=False, multi_panel=True, show=False, @@ -115,11 +215,18 @@ def main(): # Generate quality control plots generate_violin_plots( - adata, args.sample_field, args.percent_mito_field, format=args.output_format + adata, args.sample_field, args.percent_mito_field, + args.percent_ribo_field, format=args.output_format + ) + generate_scatter_plot( + adata, + args.sample_field, + percent_mito_field=args.percent_mito_field, ) generate_scatter_plot( adata, args.sample_field, + y='log1p_n_genes_by_counts', percent_mito_field=args.percent_mito_field, ) if args.doublet_score_field in adata.obs.columns: @@ -130,8 +237,24 @@ def main(): format=args.output_format, ) else: - print("Doublet score field provided not in adata.obs.columns, skipping plot.") + print( + "Doublet score field provided not in adata.obs.columns, " + "skipping plot." + ) generate_complexity_plot(adata, args.sample_field, format=args.output_format) + + if args.embedding: + generate_embedding_plot( + adata, + fields=fields + [args.sample_field, 'auto_diagnosis'], + embedding=args.embedding, + format=args.output_format, + ) + + generate_barplot(adata[~healthy_cells], + groups_field=args.sample_field, + props_field='auto_diagnosis', + figure_path='diagnosis_barplot.pdf', + topic_for_title=f"(Total Healthy/Unhealthy cells: {healthy_count}/{adata.n_obs - healthy_count})") # generate_scatter_by_sample( # adata, # sample_field=args.sample_field, @@ -140,10 +263,69 @@ def main(): # ) +def generate_barplot( + adata, groups_field, props_field, figure_path=None, topic_for_title=None +): + """ + Generate a proportional bar plot from an AnnData object. + + Parameters: + adata (AnnData): The input AnnData object containing the data to plot. + groups_field (str): The column in adata.obs to group the data by. + props_field (str): The column in adata.obs to plot as proportions. + figure_path (str, optional): The path to save the generated figure. If not provided, the figure is not saved. + topic_for_title (str, optional): The topic to be used in the figure title, goes after {props_field} proportion of {topic_for_title} per {groups_field}. + + Returns: + matplotlib.figure.Figure: The generated bar plot. + """ + props_plot_data = adata.obs[[groups_field, props_field]] + # props_plot_data[groups_field] = props_plot_data[groups_field].cat.reorder_categories(['control', '2 days', '7 days', '10 days', '14 days']) + # make a 100% stacked bar plot of props_plot_data, plotting phase counts grouped by cell_line_persister + + grouped = props_plot_data.groupby([groups_field, props_field]).size().unstack() + # proportions = grouped.div(grouped.sum(axis=1), axis=0) + colors = plt.rcParams['axes.prop_cycle'].by_key()['color'] + plt.gca().set_prop_cycle(color=colors[2:5]) + grouped.plot(kind="bar", stacked=False, figsize=(8, 6)) + if topic_for_title is not None: + plt.title(f"{props_field} cells {topic_for_title}\nper {groups_field}") + else: + plt.title(f"{props_field} cells\nper {groups_field}") + plt.xlabel(groups_field) + # plt.xticks(rotation=45) + plt.ylabel("Number of cells") + + # save plot to PDF file + if figure_path is not None: + plt.savefig(figure_path, bbox_inches="tight") + return plt.figure() + + +def generate_embedding_plot( + adata, + fields, + embedding, + format="pdf" + ): + # Embedding plot + plt.figure() + sc.pl.embedding( + adata, + basis=embedding, + color=fields, + show=False, + ncols=1 + ) + plt.savefig(f"embedding_plots.{format}", bbox_inches="tight") + plt.close() + + def generate_violin_plots( adata, sample_field, percent_mito_field="percent_mito", + percent_ribo_field="percent_ribo", format="pdf", gene_symbols_field="gene_symbols", ): @@ -171,7 +353,7 @@ def generate_violin_plots( groupby=sample_field, ax=ax, title="Number of Genes per Cell (Separated by Sample)", - show=False + show=False, # show=True, # save="_n_genes_per_cell", ) @@ -187,20 +369,35 @@ def generate_violin_plots( percent_mito_field, groupby=sample_field, ax=ax, - title="Percentage of Mitochondrial Genes per Cell (Separated by Sample)", + title="Percentage of Mitochondrial " + "Genes per Cell (Separated by Sample)", show=False, ) ax.set_xticklabels(ax.get_xticklabels(), rotation=45, ha="right") plt.savefig(f"percent_mito_per_cell.{format}", bbox_inches="tight") plt.close() + # Percentage of ribosomal genes per cell + plt.figure() + ax = plt.gca() + sc.pl.violin( + adata, + percent_ribo_field, + groupby=sample_field, + ax=ax, + title="Percentage of Ribosomal " + "Genes per Cell (Separated by Sample)", + show=False, + ) + ax.set_xticklabels(ax.get_xticklabels(), rotation=45, ha="right") + plt.savefig(f"percent_ribo_per_cell.{format}", bbox_inches="tight") + plt.close() + # highest expressed genes per cell plt.figure() ax = sc.pl.highest_expr_genes( adata, n_top=30, gene_symbols=gene_symbols_field, show=False ) # set title of ax - ax.set_title(f"Highest expressed genes per cell (Separated by Sample)") + ax.set_title("Highest expressed genes per cell\nby Sample") plt.savefig(f"highest_expr_genes.{format}", bbox_inches="tight") plt.close() @@ -223,6 +420,7 @@ def generate_violin_plots( def generate_scatter_plot( adata, sample_field, + y="n_genes", percent_mito_field="percent_mito", ): # Scatter plot of UMIs vs genes detected @@ -230,10 +428,10 @@ def generate_scatter_plot( sc.pl.scatter( adata, x="n_counts", - y="n_genes", + y=y, color=sample_field, - title="UMIs vs Genes Detected (Separated by Sample)", - save="_umi_vs_genes_detected", + title="UMIs vs Genes Detected (by Sample)", + save=f"_umi_vs_{y}_detected", show=False, ) plt.close() @@ -243,14 +441,26 @@ def generate_scatter_plot( sc.pl.scatter( adata, x="n_counts", - y="n_genes", + y=y, color=percent_mito_field, - title="UMIs vs Genes Detected (Colored by Mitochondrial Gene Ratio)", - save="_umi_vs_genes_detected_colored_by_mito", + title="UMIs vs Genes Detected (by Mitochondrial Gene Ratio)", + save=f"_umi_vs_{y}_detected_colored_by_mito", show=False, ) plt.close() + plt.figure() + sc.pl.scatter( + adata, + x='n_counts', + y=y, + color='auto_diagnosis', + title="UMIs vs Genes Detected (by Mitochondrial Gene Ratio)", + save=f"_umi_vs_{y}_detected_colored_by_auto_diagnosis", + show=False + ) + plt.close() + def generate_scatter_by_sample( adata, sample_field, percent_mito_field="percent_mito", format="pdf"