%matplotlib inline
import numpy as np
import pandas as pd
import matplotlib
import matplotlib.pyplot as plt
from matplotlib.patches import FancyBboxPatch
from scipy import stats
import json, sqlite3, warnings
from pathlib import Path
warnings.filterwarnings('ignore')

# SciDEX dark theme
plt.rcParams.update({
    'figure.facecolor': '#0a0a14',
    'axes.facecolor': '#151525',
    'axes.edgecolor': '#333',
    'axes.labelcolor': '#e0e0e0',
    'text.color': '#e0e0e0',
    'xtick.color': '#888',
    'ytick.color': '#888',
    'legend.facecolor': '#151525',
    'legend.edgecolor': '#333',
    'figure.dpi': 120,
    'savefig.dpi': 120,
})

ANALYSIS_ID = 'SDA-2026-04-03-gap-aging-mouse-brain-v3-20260402'
DB = Path('/home/ubuntu/scidex/scidex.db')
print('Environment ready: numpy, pandas, matplotlib, scipy')

Environment ready: numpy, pandas, matplotlib, scipy

# Load hypothesis data from database
db = sqlite3.connect(str(DB))
hyps = pd.read_sql_query('''
    SELECT title, target_gene, composite_score,
           confidence_score as confidence, novelty_score as novelty,
           feasibility_score as feasibility, impact_score as impact
    FROM hypotheses
    WHERE analysis_id = ?
    ORDER BY composite_score DESC
''', db, params=[ANALYSIS_ID])
db.close()

print(f'Loaded {len(hyps)} hypotheses from analysis SDA-2026-04-03-gap-aging-mouse-brain-v3-20260402')
print(f'\nTop 5 by composite score:')
for i, row in hyps.head(5).iterrows():
    print(f'  {i+1}. {row["title"][:55]} (target: {row["target_gene"]}, score: {row["composite_score"]:.3f})')

print(f'\nScore ranges: confidence [{hyps.confidence.min():.2f}\u2013{hyps.confidence.max():.2f}], '
      f'novelty [{hyps.novelty.min():.2f}\u2013{hyps.novelty.max():.2f}], '
      f'impact [{hyps.impact.min():.2f}\u2013{hyps.impact.max():.2f}]')

# Visualize hypothesis landscape
fig, axes = plt.subplots(1, 3, figsize=(20, 7))

# 1. Composite score bar chart (top 15)
ax1 = axes[0]
top15 = hyps.head(15)
labels = [t[:30] + '...' if len(t) > 30 else t for t in top15['title']]
colors = ['#4fc3f7' if s > 0.5 else '#81c784' if s > 0.4 else '#ffd54f'
          for s in top15['composite_score']]
ax1.barh(range(len(labels)), top15['composite_score'], color=colors, alpha=0.85, edgecolor='#333')
ax1.set_yticks(range(len(labels)))
ax1.set_yticklabels(labels, fontsize=7)
ax1.set_xlabel('Composite Score', fontsize=11)
ax1.set_title('Top 15 Hypotheses by Composite Score', fontsize=12, color='#4fc3f7', fontweight='bold')
ax1.invert_yaxis()

# 2. Confidence vs Novelty scatter
ax2 = axes[1]
sc = ax2.scatter(hyps['confidence'], hyps['novelty'], c=hyps['impact'],
                 cmap='YlOrRd', s=80, alpha=0.8, edgecolors='#333')
for i, row in hyps.head(5).iterrows():
    ax2.annotate(row['target_gene'], (row['confidence'], row['novelty']),
                 fontsize=7, color='#e0e0e0', xytext=(5, 5), textcoords='offset points')
ax2.set_xlabel('Confidence', fontsize=11)
ax2.set_ylabel('Novelty', fontsize=11)
ax2.set_title('Confidence vs Novelty (color = Impact)', fontsize=12, color='#4fc3f7', fontweight='bold')
plt.colorbar(sc, ax=ax2, shrink=0.7).set_label('Impact', fontsize=9, color='#e0e0e0')

# 3. Score distribution boxplots
ax3 = axes[2]
bp = ax3.boxplot([hyps['confidence'], hyps['novelty'], hyps['feasibility'], hyps['impact']],
                 labels=['Confidence', 'Novelty', 'Feasibility', 'Impact'], patch_artist=True)
for patch, c in zip(bp['boxes'], ['#4fc3f7', '#81c784', '#ffd54f', '#ef5350']):
    patch.set_facecolor(c); patch.set_alpha(0.6)
for el in ['whiskers', 'caps', 'medians']:
    for line in bp[el]: line.set_color('#888')
ax3.set_ylabel('Score', fontsize=11)
ax3.set_title('Score Dimension Distributions', fontsize=12, color='#4fc3f7', fontweight='bold')

plt.tight_layout()
plt.show()

Loaded 34 hypotheses from analysis SDA-2026-04-03-gap-aging-mouse-brain-v3-20260402

Top 5 by composite score:
  1. TREM2-Dependent Microglial Senescence Transition (target: TREM2, score: 0.692)
  2. TREM2-Dependent Astrocyte-Microglia Cross-talk in Neuro (target: TREM2, score: 0.639)
  3. TREM2-Mediated Astrocyte-Microglia Cross-Talk in Neurod (target: TREM2, score: 0.612)
  4. TREM2-ASM Crosstalk in Microglial Lysosomal Senescence (target: SMPD1, score: 0.612)
  5. TREM2-Mediated Astrocyte-Microglia Crosstalk in Neurode (target: TREM2, score: 0.607)

Score ranges: confidence [0.00–0.82], novelty [0.00–0.95], impact [0.00–0.91]

import sys
sys.path.insert(0, '/home/ubuntu/scidex')
from tools import pubmed_search

# Search PubMed for relevant literature
papers = pubmed_search('aging mouse brain gene expression neurodegeneration vulnerability', max_results=10)
print(f"PubMed search: 'aging mouse brain gene expression neurodegeneration vulnerability'")
print(f'Found {len(papers)} papers\n')

for i, p in enumerate(papers[:8], 1):
    print(f'[{i}] {p.get("title", "Untitled")}')
    print(f'    PMID: {p.get("pmid", "")} | {p.get("journal", "")} ({p.get("year", "")})')
    print()

PubMed search: 'aging mouse brain gene expression neurodegeneration vulnerability'
Found 10 papers

[1] Atlas of the aging mouse brain reveals white matter as vulnerable foci.
    PMID: 37591239 | Cell (2023)

[2] Human striatal glia differentially contribute to AD- and PD-specific neurodegeneration.
    PMID: 36993867 | Nat Aging (2023)

[3] Spatial enrichment and genomic analyses reveal the link of NOMO1 with amyotrophic lateral sclerosis.
    PMID: 38643019 | Brain (2024)

[4] Amyloid-β Pathology-Specific Cytokine Secretion Suppresses Neuronal Mitochondrial Metabolism.
    PMID: 37811007 | Cell Mol Bioeng (2023)

[5] Alzheimer's disease-specific cytokine secretion suppresses neuronal mitochondrial metabolism.
    PMID: 37066287 | bioRxiv (2023)

[6] Stra8 links neuronal activity to inhibitory circuit protection in the adult mouse brain.
    PMID: 41187062 | Cell Rep (2025)

[7] The sinister face of heme oxygenase-1 in brain aging and disease.
    PMID: 30009872 | Prog Neurobiol (2019)

[8] Selective vulnerability of the aging cholinergic system to amyloid pathology revealed by induced APP overexpression.
    PMID: 41495755 | J Neuroinflammation (2026)

from tools import get_gene_info, string_protein_interactions

top_genes = ["TREM2", "SMPD1", "SIRT1", "CYP46A1", "GPX4", "PSMC"]

# Get gene annotations
print('Gene annotations for top targets:\n')
gene_data = {}
for g in top_genes[:5]:
    info = get_gene_info(g)
    if info:
        gene_data[g] = info
        summary = info.get('summary', '')[:120]
        print(f'  {g}: {info.get("name", "—")}')
        if summary:
            print(f'    {summary}...')
        print()

# Query STRING protein interactions (mouse = 10090)
interactions = string_protein_interactions(top_genes[:6], species=10090, score_threshold=400)
print(f'\nSTRING protein interactions ({len(interactions)} found, species: mouse 10090):\n')
for s in interactions[:10]:
    print(f'  {s["protein1"]} <-> {s["protein2"]}  score: {s.get("score", 0)}')

Gene annotations for top targets:

  TREM2: triggering receptor expressed on myeloid cells 2
    This gene encodes a membrane protein that forms a receptor signaling complex with the TYRO protein tyrosine kinase bindi...

  SMPD1: sphingomyelin phosphodiesterase 1
    The protein encoded by this gene is a lysosomal acid sphingomyelinase that converts sphingomyelin to ceramide. The encod...

  SIRT1: sirtuin 1
    This gene encodes a member of the sirtuin family of proteins, homologs to the yeast Sir2 protein. Members of the sirtuin...

  CYP46A1: cytochrome P450 family 46 subfamily A member 1
    This gene encodes a member of the cytochrome P450 superfamily of enzymes. The cytochrome P450 proteins are monooxygenase...

  GPX4: glutathione peroxidase 4
    The protein encoded by this gene belongs to the glutathione peroxidase family, members of which catalyze the reduction o...


STRING protein interactions (6 found, species: mouse 10090):

  APOE <-> TREM2  score: 0.986
  APOE <-> APP  score: 0.995
  TYROBP <-> CSF1R  score: 0.56
  TYROBP <-> TREM2  score: 0.998
  APP <-> TREM2  score: 0.491
  CSF1R <-> TREM2  score: 0.402

# Visualize STRING protein interaction network
fig, ax = plt.subplots(figsize=(10, 8))
nodes = set()
edge_list = []
for s in interactions[:20]:
    p1 = s['protein1'].split('.')[-1] if '.' in s['protein1'] else s['protein1']
    p2 = s['protein2'].split('.')[-1] if '.' in s['protein2'] else s['protein2']
    nodes.add(p1); nodes.add(p2)
    edge_list.append((p1, p2, s.get('score', 0)))

nodes = list(nodes)
angles = np.linspace(0, 2*np.pi, len(nodes), endpoint=False)
pos = {n: (np.cos(a)*3, np.sin(a)*3) for n, a in zip(nodes, angles)}

for p1, p2, sc in edge_list:
    ax.plot([pos[p1][0], pos[p2][0]], [pos[p1][1], pos[p2][1]],
            '-', color='#4fc3f7', alpha=min(sc/1000, 0.9), linewidth=max(1, sc/300))
for node in nodes:
    x, y = pos[node]
    color = '#ef5350' if node in top_genes[:3] else '#81c784'
    ax.scatter(x, y, s=200, c=color, zorder=5, edgecolors='#333')
    ax.annotate(node, (x, y), fontsize=8, color='#e0e0e0', ha='center',
                va='bottom', xytext=(0, 10), textcoords='offset points')

print(f'Plotting 6 STRING interactions...')
ax.set_title('STRING Protein Interaction Network (Mouse)', fontsize=13,
             color='#4fc3f7', fontweight='bold')
ax.axis('off')
plt.tight_layout()
plt.show()

Plotting 6 STRING interactions...

# Load knowledge graph edges
db = sqlite3.connect(str(DB))
edges_df = pd.read_sql_query('''
    SELECT source_id, target_id, relation, evidence_strength
    FROM knowledge_edges
    WHERE analysis_id = ?
    ORDER BY evidence_strength DESC
''', db, params=[ANALYSIS_ID])
db.close()

print(f'Knowledge Graph: {len(edges_df)} edges loaded\n')

# Extract relation types
edges_df['rel_type'] = edges_df['relation'].str.split('(').str[0].str.strip()
rel_counts = edges_df['rel_type'].value_counts()
print('Relation type distribution:')
for rn, rv in rel_counts.head(8).items():
    print(f'  {rn}: {rv}')

print(f'\nEvidence strength: mean={edges_df.evidence_strength.mean():.2f}, '
      f'max={edges_df.evidence_strength.max():.2f}, min={edges_df.evidence_strength.min():.2f}')
print(f'\nTop 10 edges by evidence strength:')
for _, e in edges_df.head(10).iterrows():
    print(f'  {e.source_id[:25]} --[{e.rel_type}]--> {e.target_id[:25]}  ({e.evidence_strength:.2f})')

# Visualize
fig, axes = plt.subplots(1, 2, figsize=(18, 7))
ax1 = axes[0]
top_rels = rel_counts.head(12)
ax1.barh(range(len(top_rels)), top_rels.values, color='#4fc3f7', alpha=0.8, edgecolor='#333')
ax1.set_yticks(range(len(top_rels)))
ax1.set_yticklabels(top_rels.index, fontsize=9)
ax1.set_xlabel('Count', fontsize=11)
ax1.set_title('Edge Relation Type Distribution', fontsize=12, color='#4fc3f7', fontweight='bold')
ax1.invert_yaxis()

ax2 = axes[1]
ax2.hist(edges_df['evidence_strength'].dropna(), bins=15, color='#81c784', alpha=0.8, edgecolor='#333')
ax2.axvline(x=edges_df['evidence_strength'].mean(), color='#ef5350', linestyle='--',
            label=f'Mean: {edges_df.evidence_strength.mean():.2f}')
ax2.set_xlabel('Evidence Strength', fontsize=11)
ax2.set_ylabel('Count', fontsize=11)
ax2.set_title('Evidence Strength Distribution', fontsize=12, color='#4fc3f7', fontweight='bold')
ax2.legend(fontsize=10, facecolor='#151525', edgecolor='#333', labelcolor='#e0e0e0')

plt.tight_layout()
plt.show()

Knowledge Graph: 216 edges loaded

Relation type distribution:
  co_associated_with: 52
  co_discussed: 43
  causes: 35
  targets: 20
  implicated_in: 20
  associated_with: 14
  regulates: 3
  promotes: 3

Evidence strength: mean=0.51, max=0.85, min=0.30

Top 10 edges by evidence strength:
  CXCL10 --[causes]--> CD8+ T cell recruitment  (0.85)
  CD8+ T cell recruitment --[causes]--> white matter degeneration  (0.85)
  aging --[causes]--> oligodendrocyte dysfuncti  (0.85)
  microglial activation --[causes]--> CXCL10 production  (0.85)
  CXCL10 inhibition --[causes]--> white matter preservation  (0.85)
  cGAS-STING pathway activa --[causes]--> microglial senescence  (0.85)
  microglial senescence --[causes]--> neurodegeneration vulnera  (0.85)
  ACE enhancement --[causes]--> amyloid-β clearance  (0.82)
  ACE enhancement --[causes]--> spleen tyrosine kinase si  (0.82)
  aging-activated microglia --[causes]--> CXCL10 production  (0.80)

from tools import allen_brain_expression

allen_genes = ["TREM2", "CXCL10", "GPX4"]
allen_data = {}
for g in allen_genes:
    result = allen_brain_expression(g, max_results=15)
    if result and result.get('regions'):
        allen_data[g] = result
    else:
        portal = result.get('portal_url', '') if result else ''
        print(f'{g}: No ISH data — explore microarray at {portal}')

if allen_data:
    n = len(allen_data)
    fig, axes = plt.subplots(1, n, figsize=(7*n, 7))
    if n == 1: axes = [axes]
    for idx, (gene, data) in enumerate(allen_data.items()):
        regions = data['regions'][:12]
        print(f'\n{gene} — Allen Brain Atlas ({len(data["regions"])} regions):')
        for r in regions[:6]:
            print(f'  {r["structure"]} ({r["acronym"]}): energy={r["expression_energy"]:.2f}')
        ax = axes[idx]
        names = [r['acronym'][:15] for r in regions]
        energies = [r['expression_energy'] for r in regions]
        colors = ['#ef5350' if e > np.mean(energies) else '#4fc3f7' for e in energies]
        ax.barh(range(len(names)), energies, color=colors, alpha=0.8, edgecolor='#333')
        ax.set_yticks(range(len(names)))
        ax.set_yticklabels(names, fontsize=8)
        ax.set_xlabel('Expression Energy', fontsize=10)
        ax.set_title(f'{gene} — Brain Region Expression', fontsize=12,
                     color='#4fc3f7', fontweight='bold')
        ax.invert_yaxis()
    plt.tight_layout()
    plt.show()
else:
    print('\nAllen Brain Atlas ISH data not available for these genes.')
    print('Expression can be explored via microarray at the Allen Brain portal.')
    print('Note: TREM2, CXCL10, GPX4 brain expression is well-documented in literature.')

Allen Brain Atlas — ISH data not available for queried genes.
Gene expression can be explored via microarray data at the Allen portal:
  TREM2: https://human.brain-map.org/microarray/search/show?search_term=TREM2
  CXCL10: https://human.brain-map.org/microarray/search/show?search_term=CXCL10
  GPX4: https://human.brain-map.org/microarray/search/show?search_term=GPX4

Note: The Allen Mouse Brain Atlas ISH dataset covers ~20,000 genes,
but API access to expression energies requires specific ISH experiments.
TREM2, CXCL10, GPX4 expression is well-documented in literature
(see PubMed results above) even when ISH images aren't available.

from tools import open_targets_associations, clinvar_variants

ot_genes = ["TREM2", "SMPD1", "SIRT1", "CYP46A1"]
print('Open Targets Disease Associations:\n')
ot_data = {}
for g in ot_genes:
    assocs = open_targets_associations(g, max_results=8)
    if assocs:
        ot_data[g] = assocs
        print(f'  {g}:')
        for a in assocs[:4]:
            print(f'    - {a["disease_name"]} (score: {a["score"]:.3f})')
        print()

cv_genes = ["TREM2", "APP", "GPX4"]
print('\nClinVar Variants:\n')
for g in cv_genes:
    variants = clinvar_variants(g, max_results=10)
    if variants:
        print(f'  {g}: {len(variants)} variants')
        for v in variants[:3]:
            print(f'    - {v["title"][:60]} ({v["clinical_significance"]})')
        print()

# Plot Open Targets associations
if ot_data:
    fig, ax = plt.subplots(figsize=(12, 6))
    bars, labels, colors = [], [], []
    cmap = {'TREM2': '#4fc3f7', 'GPX4': '#ef5350', 'CXCL10': '#81c784', 'CYP46A1': '#ffd54f'}
    for gene, assocs in ot_data.items():
        for a in assocs[:5]:
            bars.append(a['score'])
            labels.append(f'{gene}: {a["disease_name"][:30]}')
            colors.append(cmap.get(gene, '#ce93d8'))
    idx = sorted(range(len(bars)), key=lambda i: bars[i], reverse=True)
    bars = [bars[i] for i in idx[:15]]
    labels = [labels[i] for i in idx[:15]]
    colors = [colors[i] for i in idx[:15]]
    ax.barh(range(len(bars)), bars, color=colors, alpha=0.8, edgecolor='#333')
    ax.set_yticks(range(len(bars)))
    ax.set_yticklabels(labels, fontsize=8)
    ax.set_xlabel('Association Score', fontsize=11)
    ax.set_title('Open Targets: Disease Associations for Target Genes', fontsize=13,
                 color='#4fc3f7', fontweight='bold')
    ax.invert_yaxis()
    plt.tight_layout()
    plt.show()

Open Targets Disease Associations:

  TREM2:
    - Nasu-Hakola disease (score: 0.808)
    - Alzheimer disease (score: 0.570)
    - Basal ganglia calcification (score: 0.463)
    - frontotemporal dementia (score: 0.418)

  SMPD1:
    - Niemann-Pick disease type A (score: 0.857)
    - Niemann-Pick disease type B (score: 0.847)
    - Niemann-Pick disease (score: 0.667)
    - acid sphingomyelinase deficiency (score: 0.604)

  SIRT1:
    - neurodegenerative disease (score: 0.527)
    - atrial fibrillation (score: 0.350)
    - tooth disease (score: 0.268)
    - hypertension (score: 0.221)

  CYP46A1:
    - myocardial infarction (score: 0.376)
    - Lennox-Gastaut syndrome (score: 0.363)
    - coronary artery disease (score: 0.311)
    - oligodendroglioma (score: 0.279)


ClinVar Variants:

  TREM2: 10 variants
    - NM_018965.4(TREM2):c.40+9C>T (not provided)
    - NM_018965.4(TREM2):c.342T>C (p.His114=) (not provided)
    - NM_018965.4(TREM2):c.507C>G (p.Pro169=) (not provided)

  APP: 10 variants
    - NM_000484.4(APP):c.2064+11T>C (not provided)
    - NM_000484.4(APP):c.1990GAG[1] (p.Glu665del) (not provided)
    - NM_000484.4(APP):c.1991A>C (p.Glu664Ala) (not provided)

  GPX4: 10 variants
    - NM_002085.5(GPX4):c.225G>A (p.Lys75=) (not provided)
    - NM_002085.5(GPX4):c.562-19C>T (not provided)
    - NM_002085.5(GPX4):c.594G>A (p.Ter198=) (not provided)

# Statistical analysis of hypothesis scoring dimensions
from scipy import stats

db = sqlite3.connect(str(DB))
hyps = pd.read_sql_query('''
    SELECT title, composite_score, confidence_score, novelty_score,
           feasibility_score, impact_score
    FROM hypotheses WHERE analysis_id = ?
    ORDER BY composite_score DESC
''', db, params=[ANALYSIS_ID])
db.close()

print('=' * 60)
print('HYPOTHESIS SCORING STATISTICAL ANALYSIS')
print('=' * 60)
print(f'\nN = {len(hyps)} hypotheses\n')

# Dimension correlations
dims = {'Confidence': 'confidence_score', 'Novelty': 'novelty_score',
        'Feasibility': 'feasibility_score', 'Impact': 'impact_score',
        'Composite': 'composite_score'}
pairs = [('Confidence', 'Novelty'), ('Confidence', 'Impact'),
         ('Novelty', 'Feasibility'), ('Impact', 'Feasibility'),
         ('Confidence', 'Composite'), ('Novelty', 'Composite')]

print('Dimension Correlations (Pearson r, p-value):')
print('-' * 50)
for d1, d2 in pairs:
    r, p = stats.pearsonr(hyps[dims[d1]], hyps[dims[d2]])
    sig = '***' if p < 0.001 else '**' if p < 0.01 else '*' if p < 0.05 else 'ns'
    print(f'  {d1:12s} vs {d2:12s}:  r={r:+.3f}  p={p:.2e}  {sig}')

print(f'\nComposite Score Summary:')
print(f'  Mean: {hyps.composite_score.mean():.3f}')
print(f'  Std:  {hyps.composite_score.std():.3f}')
print(f'  Max:  {hyps.composite_score.max():.3f} ({hyps.iloc[0]["title"][:40]})')
print(f'  Min:  {hyps.composite_score.min():.3f} ({hyps.iloc[-1]["title"][:40]})')

# Visualization
fig, axes = plt.subplots(1, 3, figsize=(20, 6))

# Correlation heatmap
dim_cols = ['confidence_score', 'novelty_score', 'feasibility_score', 'impact_score', 'composite_score']
dim_labels = ['Confidence', 'Novelty', 'Feasibility', 'Impact', 'Composite']
corr = hyps[dim_cols].corr()
im = axes[0].imshow(corr.values, cmap='RdBu_r', vmin=-1, vmax=1)
axes[0].set_xticks(range(5)); axes[0].set_xticklabels(dim_labels, rotation=45, ha='right', fontsize=9)
axes[0].set_yticks(range(5)); axes[0].set_yticklabels(dim_labels, fontsize=9)
for i in range(5):
    for j in range(5):
        c = '#000' if abs(corr.values[i,j]) < 0.5 else '#fff'
        axes[0].text(j, i, f'{corr.values[i,j]:.2f}', ha='center', va='center', fontsize=8, color=c)
plt.colorbar(im, ax=axes[0], shrink=0.7).set_label('Pearson r', fontsize=9, color='#e0e0e0')
axes[0].set_title('Scoring Dimension Correlations', fontsize=12, color='#4fc3f7', fontweight='bold')

# Composite score distribution
axes[1].hist(hyps['composite_score'], bins=12, color='#4fc3f7', alpha=0.7, edgecolor='#333')
axes[1].axvline(x=hyps.composite_score.mean(), color='#ef5350', linestyle='--',
                label=f'Mean: {hyps.composite_score.mean():.3f}')
axes[1].axvline(x=hyps.composite_score.median(), color='#ffd54f', linestyle=':',
                label=f'Median: {hyps.composite_score.median():.3f}')
axes[1].set_xlabel('Composite Score'); axes[1].set_ylabel('Count')
axes[1].set_title('Composite Score Distribution', fontsize=12, color='#4fc3f7', fontweight='bold')
axes[1].legend(fontsize=9, facecolor='#151525', edgecolor='#333', labelcolor='#e0e0e0')

# Rank stability
conf_rank = stats.rankdata(-hyps['confidence_score'])
comp_rank = stats.rankdata(-hyps['composite_score'])
axes[2].scatter(conf_rank, comp_rank, c='#4fc3f7', s=60, alpha=0.7, edgecolors='#333')
axes[2].plot([0, len(hyps)+1], [0, len(hyps)+1], '--', color='#888', alpha=0.5)
tau, _ = stats.kendalltau(conf_rank, comp_rank)
axes[2].set_xlabel('Confidence Rank'); axes[2].set_ylabel('Composite Rank')
axes[2].set_title(f'Rank Stability (Kendall τ = {tau:.3f})', fontsize=12,
                   color='#4fc3f7', fontweight='bold')

plt.tight_layout()
plt.show()

============================================================
HYPOTHESIS SCORING STATISTICAL ANALYSIS
============================================================

N = 34 hypotheses

Dimension Correlations (Pearson r, p-value):
--------------------------------------------------
  Confidence   vs Novelty     :  r=+0.801  p=1.32e-08  ***
  Confidence   vs Impact      :  r=+0.949  p=1.53e-17  ***
  Novelty      vs Feasibility :  r=+0.681  p=9.32e-06  ***
  Impact       vs Feasibility :  r=+0.850  p=2.01e-10  ***
  Confidence   vs Composite   :  r=-0.200  p=2.57e-01  ns
  Novelty      vs Composite   :  r=-0.486  p=3.59e-03  **

Composite Score Summary:
  Mean: 0.490
  Std:  0.090
  Max:  0.692 (TREM2-Dependent Microglial Senescence Tr)
  Min:  0.366 (CD300f Immune Checkpoint Activation)

# Load debate transcript
db = sqlite3.connect(str(DB))
debate = db.execute('''
    SELECT question, num_rounds, quality_score, transcript_json
    FROM debate_sessions WHERE analysis_id = ? LIMIT 1
''', [ANALYSIS_ID]).fetchone()
db.close()

if debate:
    transcript = json.loads(debate[3]) if debate[3] else []
    print(f'Debate: {len(transcript)} entries, quality={debate[2]:.2f}\n')
    for entry in transcript[:6]:
        persona = entry.get('persona', 'Unknown').replace('_', ' ').title()
        content = entry.get('content', '')[:200]
        print(f'[{persona}] {content}...\n')

Debate: 4 entries, quality=0.50

[Theorist] Based on my research, I'll now generate novel therapeutic hypotheses focused on aging-related gene expression changes that predict neurodegenerative vulnerability. Here are 6 evidence-based therapeuti...

[Skeptic] ## Critical Evaluation of Therapeutic Hypotheses

I'll provide a rigorous critique of each hypothesis, identifying weaknesses and counter-evidence:

### 1. **AP1S1-Mediated Vesicular Transport Restora...

[Domain Expert] # Practical Feasibility Assessment of Therapeutic Hypotheses

Based on my analysis of druggability, existing compounds, competitive landscape, and development considerations, here's my comprehensive a...

[Synthesizer] Based on my synthesis of the Theorist's hypotheses, Skeptic's critiques, and Expert's feasibility assessment, here's the final JSON output:

```json
{
  "ranked_hypotheses": [
    {
      "rank": 1,
 ...