Statistical Significance Annotation

from scipy import stats

# Two-group comparison
stat, pval = stats.mannwhitneyu(group_a, group_b, alternative='two-sided')
# or for normal data:
stat, pval = stats.ttest_ind(group_a, group_b)

# Multi-group: ANOVA + post-hoc
from scipy.stats import f_oneway
stat, pval_anova = f_oneway(group_a, group_b, group_c)

# Post-hoc pairwise (if ANOVA significant)
from itertools import combinations
from statsmodels.stats.multitest import multipletests

pairs = list(combinations(["A", "B", "C"], 2))
groups = {"A": group_a, "B": group_b, "C": group_c}
raw_pvals = []
for g1, g2 in pairs:
    _, p = stats.mannwhitneyu(groups[g1], groups[g2], alternative='two-sided')
    raw_pvals.append(p)

# Adjust for multiple comparisons
rejected, adj_pvals, _, _ = multipletests(raw_pvals, method='fdr_bh')

def add_significance_bracket(ax, x1, x2, y, p_value, dh=0.02, barh=0.015, fontsize=11):
    """Draw a significance bracket with asterisk notation between two x positions.

    Args:
        ax: matplotlib Axes object.
        x1, x2: x-axis positions of the two groups (0-indexed).
        y: y-coordinate for the bracket (top of bracket line).
        p_value: p-value for the comparison.
        dh: vertical offset above bracket for the text (in axes fraction).
        barh: height of the bracket tips (in axes fraction).
        fontsize: font size for the asterisk text.
    """
    asterisk = pvalue_to_asterisk(p_value)

    # Draw bracket: two tips and a connecting line
    ax.plot([x1, x1, x2, x2], [y - barh, y, y, y - barh],
            lw=1.2, color='black')
    # Place asterisk text centered above the bracket
    ax.text((x1 + x2) / 2, y + dh, asterisk,
            ha='center', va='bottom', fontsize=fontsize, fontweight='bold')

import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np

# Example: box plot with significance annotation
fig, ax = plt.subplots(figsize=(6, 5))
sns.boxplot(data=df, x="group", y="value", ax=ax, palette="Set2")
sns.stripplot(data=df, x="group", y="value", ax=ax,
              color="black", alpha=0.4, size=3, jitter=True)

# Determine bracket y-position from data
y_max = df["value"].max()
y_range = df["value"].max() - df["value"].min()
offset = y_range * 0.08  # spacing between brackets

# Add brackets for each significant comparison
# pairs_with_pvals: list of (group1_idx, group2_idx, p_value)
pairs_with_pvals = [(0, 1, 0.003), (0, 2, 0.042)]

for i, (x1, x2, pval) in enumerate(pairs_with_pvals):
    bracket_y = y_max + offset * (i + 1)
    add_significance_bracket(ax, x1, x2, bracket_y, pval)

ax.set_title("Gene Expression by Treatment", fontweight='bold', fontsize=14)
ax.set_ylabel("Expression (log2 CPM)")
# Extend y-axis to fit brackets
ax.set_ylim(top=y_max + offset * (len(pairs_with_pvals) + 1.5))

plt.tight_layout()
plt.savefig("expression_comparison.png", dpi=150, bbox_inches='tight')

fig, ax = plt.subplots(figsize=(7, 5))
bar_plot = sns.barplot(data=df, x="gene", y="fold_change", hue="condition",
                       ax=ax, palette="Set2", ci="sd", capsize=0.05)

# For grouped bars, calculate x positions manually
# Each gene has multiple bars offset by group
n_groups = df["condition"].nunique()
n_genes = df["gene"].nunique()
bar_width = 0.8 / n_groups

for gene_idx in range(n_genes):
    # x positions of the two bars within this gene group
    x1 = gene_idx - bar_width / 2
    x2 = gene_idx + bar_width / 2
    # Get the max value + error for this gene
    gene_data = df[df["gene"] == df["gene"].unique()[gene_idx]]
    y_top = gene_data["fold_change"].mean() + gene_data["fold_change"].std()
    p_val = pvals_per_gene[gene_idx]  # pre-computed

    if p_val <= 0.05:  # only annotate significant results
        add_significance_bracket(ax, x1, x2, y_top + 0.1, p_val)

ax.set_title("Fold Change by Condition", fontweight='bold', fontsize=14)
plt.tight_layout()
plt.savefig("fold_change_comparison.png", dpi=150, bbox_inches='tight')