R1: Genetic Validation - GWAS of Signature Exposure¶

Reviewer Question¶

Referee #1: "The authors say in several places that the models describe clinically meaningful biological processes without giving any proof of the clinical and certainly not biological meaningfulness."

Why This Matters¶

Demonstrating genetic associations with signature exposure provides evidence that signatures have a genetic basis and capture biologically meaningful pathways.

Our Approach¶

We performed genome-wide association studies (GWAS) using signature exposure as quantitative phenotypes:

  1. Calculate Average Signature Exposure (AEX): For each individual, we compute the average signature loading over time
  2. GWAS on AEX: Test genome-wide SNPs for association with signature exposure
  3. Identify Signature-Specific Loci: Find genetic variants associated with signatures but not with individual diseases
  4. Map to Nearest Genes: Annotate significant hits with nearest genes

Key Innovation: We test genetic loading of 0 (baseline genetic effects) to identify loci that influence signature trajectories independently.

Key Findings¶

✅ Multiple genome-wide significant loci identified for each signature ✅ Signature-specific loci found that are not associated with individual diseases ✅ Biologically plausible gene associations (e.g., lipid genes for Signature 5)

================================================================================
GWAS LOCI SUMMARY
================================================================================

Total loci: 151

Signatures with loci: 16

Loci per signature:
Signature N_Loci Percentage
12 SIG5 56 37.1%
7 SIG17 19 12.6%
13 SIG7 14 9.3%
9 SIG19 11 7.3%
1 SIG1 8 5.3%
0 SIG0 6 4.0%
2 SIG10 6 4.0%
10 SIG2 6 4.0%
11 SIG3 6 4.0%
3 SIG13 5 3.3%
4 SIG14 4 2.6%
5 SIG15 4 2.6%
14 SIG8 3 2.0%
6 SIG16 1 0.7%
8 SIG18 1 0.7%
15 SIG9 1 0.7%
SIG #CHR POS UID EA OA EAF BETA SE LOG10P ... cb_start cb_end cytoband giemsa pops locus_id_chr locus_id_start locus_id_end locus_id Unnamed: 54
0 SIG0 20 42954982 20:42954982:A:G G A 0.011206 0.018610 0.003373 7.46280 ... 39759919 47823902 q12 gpos NaN 20 42454982 43454982 116 NaN
1 SIG0 21 30519457 21:30519457:T:A A T 0.157962 0.005564 0.000977 7.90813 ... -1 -1 . . NaN 21 30019457 31096186 119 NaN
2 SIG0 4 111718067 4:111718067:G:A A G 0.804755 -0.007606 0.000893 16.79370 ... 96010721 123004616 q24 gneg NaN 4 111107315 112231200 34 NaN
3 SIG0 6 160997118 6:160997118:A:T T A 0.080091 0.008127 0.001307 9.30298 ... -1 -1 . . NaN 6 159742487 162167545 47 NaN
4 SIG0 8 102490380 8:102490380:C:T T C 0.847157 0.005424 0.000986 7.41973 ... 84457765 109374423 q31 gpos NaN 8 101990380 102990380 52 NaN
5 SIG0 9 97590631 9:97590631:T:A A T 0.678148 0.004440 0.000759 8.30449 ... 96441248 108417612 q36 gneg NaN 9 97090631 98090631 57 NaN
6 SIG10 1 196652124 1:196652124:T:TA TA T 0.616718 -0.005430 0.000800 10.94850 ... 194712323 203473144 q36 gneg NaN 1 196146176 197427791 8 NaN
7 SIG10 10 124230024 10:124230024:A:C C A 0.212114 0.007976 0.000954 16.21150 ... -1 -1 . . NaN 10 123709684 124735355 68 NaN
8 SIG10 11 86400443 11:86400443:A:G G A 0.613320 -0.004536 0.000800 7.84024 ... 81728398 93518069 q23 gvar NaN 11 85899411 86900443 73 NaN
9 SIG10 15 27498832 15:27498832:G:A A G 0.164789 -0.006111 0.001049 8.24171 ... 17945541 33087091 p14 gneg NaN 15 26998832 27998832 91 NaN

10 rows × 55 columns

2. Top Loci by Signature¶

For each signature, we identify the top genetic loci (by p-value) and their nearest genes.

3. Novel Genetic Discoveries: Signature-Specific Loci¶

A key finding from our signature-based GWAS is the identification of novel loci that are genome-wide significant for composite signatures but not found in individual disease GWAS. This demonstrates the power of joint modeling to detect pleiotropic effects that are too weak to detect in single-disease analyses.

The 10 Unique Signature 5 Discoveries¶

For Signature 5 (cardiovascular/lipid), we identified 10 unique loci that are genome-wide significant (p < 5×10⁻⁸) in the signature-based GWAS but are not present in any constituent trait GWAS (Angina, MI, Hypercholesterolemia, Coronary atherosclerosis, Acute IHD, Chronic IHD) within a 1MB window. These represent novel discoveries enabled by joint modeling of multiple related diseases.

Key Examples:

  • rs1532085 (LIPC): Hepatic lipase gene involved in HDL metabolism. Not significant in any individual CV trait GWAS, but genome-wide significant (p = 3.8×10⁻⁸) in Signature 5, demonstrating distributed pleiotropic effects across multiple cardiovascular traits.
  • rs6687726 (IL6R): Interleukin-6 receptor, key inflammation pathway. Novel discovery through joint modeling.
  • rs1499813 (FNDC3B): Insulin signaling and adipogenesis. Strongest novel association (p = 1.1×10⁻¹⁰).
================================================================================
TOP 10 GENETIC LOCI PER SIGNATURE
================================================================================

SIG0 - Heart Failure/Arrhythmia (7 total loci)
  rs10455872      LPA                  p=2.75e-130 (LOG10P=129.56)
  rs6843082       PITX2                p=1.61e-17 (LOG10P=16.79)
  rs74617384      LPA                  p=4.98e-10 (LOG10P=9.30)
  rs10125609      C9orf3               p=4.96e-09 (LOG10P=8.30)
  rs12627426      MAP3K7CL             p=1.24e-08 (LOG10P=7.91)
  rs77410568      R3HDML               p=3.45e-08 (LOG10P=7.46)
  rs2509765       KB-1562D12.1         p=3.80e-08 (LOG10P=7.42)

SIG16 - Neurodegeneration (2 total loci)
  rs7412          APOE                 p=1.84e-59 (LOG10P=58.74)
  rs429358        APOE                 p=9.88e-14 (LOG10P=13.01)

SIG17 - GI/Colorectal (29 total loci)
  rs1333042       CDKN2B-AS1           p=9.02e-100 (LOG10P=99.04)
  rs4977575       CDKN2B-AS1           p=1.18e-20 (LOG10P=19.93)
  rs58658771      GREM1                p=1.32e-18 (LOG10P=17.88)
  rs687621        RP11-430N14.4        p=1.72e-17 (LOG10P=16.76)
  rs9275218       HLA-DQB1             p=5.53e-17 (LOG10P=16.26)
  rs4939567       SMAD7                p=5.40e-16 (LOG10P=15.27)
  rs6121558       RPS21                p=3.24e-14 (LOG10P=13.49)
  rs3184504       SH2B3                p=1.13e-13 (LOG10P=12.95)
  rs10774625      ATXN2                p=1.40e-12 (LOG10P=11.85)
  rs16888589      EIF3H                p=5.19e-12 (LOG10P=11.28)

SIG5 - Cardiovascular/Lipid (78 total loci)
  rs10455872      LPA                  p=2.75e-130 (LOG10P=129.56)
  rs1333042       CDKN2B-AS1           p=9.02e-100 (LOG10P=99.04)
  rs7412          APOE                 p=1.84e-59 (LOG10P=58.74)
  rs11887534      ABCG8                p=1.27e-55 (LOG10P=54.90)
  rs12740374      CELSR2               p=1.66e-42 (LOG10P=41.78)
  rs138294113     LDLR                 p=5.90e-39 (LOG10P=38.23)
  rs9349379       PHACTR1              p=6.50e-30 (LOG10P=29.19)
  rs11591147      PCSK9                p=7.50e-25 (LOG10P=24.12)
  rs4977575       CDKN2B-AS1           p=1.18e-20 (LOG10P=19.93)
  rs2351524       NBEAL1               p=2.30e-20 (LOG10P=19.64)

SIG7 - Hypertension/Vascular (23 total loci)
  rs9275218       HLA-DQB1             p=5.53e-17 (LOG10P=16.26)
  rs9272451       HLA-DQA1             p=4.94e-15 (LOG10P=14.31)
  rs3184504       SH2B3                p=1.13e-13 (LOG10P=12.95)
  rs10774625      ATXN2                p=1.40e-12 (LOG10P=11.85)
  rs1275977       KCNK3                p=3.06e-12 (LOG10P=11.51)
  rs7192155       CFDP1                p=4.60e-11 (LOG10P=10.34)
  rs12509595      FGF5                 p=1.57e-10 (LOG10P=9.80)
  rs2071278       NOTCH4               p=3.29e-10 (LOG10P=9.48)
  rs3806155       BTNL2                p=3.33e-10 (LOG10P=9.48)
  rs72831343      C10orf107            p=3.82e-10 (LOG10P=9.42)
Signature SNP Nearest_Gene LOG10P P_value
0 SIG0 rs10455872 LPA 129.56 2.75e-130
1 SIG0 rs6843082 PITX2 16.79 1.61e-17
2 SIG0 rs74617384 LPA 9.30 4.98e-10
3 SIG0 rs10125609 C9orf3 8.30 4.96e-09
4 SIG0 rs12627426 MAP3K7CL 7.91 1.24e-08
5 SIG0 rs77410568 R3HDML 7.46 3.45e-08
6 SIG0 rs2509765 KB-1562D12.1 7.42 3.80e-08
7 SIG16 rs7412 APOE 58.74 1.84e-59
8 SIG16 rs429358 APOE 13.01 9.88e-14
9 SIG17 rs1333042 CDKN2B-AS1 99.04 9.02e-100
10 SIG17 rs4977575 CDKN2B-AS1 19.93 1.18e-20
11 SIG17 rs58658771 GREM1 17.88 1.32e-18
12 SIG17 rs687621 RP11-430N14.4 16.76 1.72e-17
13 SIG17 rs9275218 HLA-DQB1 16.26 5.53e-17
14 SIG17 rs4939567 SMAD7 15.27 5.40e-16
15 SIG17 rs6121558 RPS21 13.49 3.24e-14
16 SIG17 rs3184504 SH2B3 12.95 1.13e-13
17 SIG17 rs10774625 ATXN2 11.85 1.40e-12
18 SIG17 rs16888589 EIF3H 11.28 5.19e-12
19 SIG5 rs10455872 LPA 129.56 2.75e-130
20 SIG5 rs1333042 CDKN2B-AS1 99.04 9.02e-100
21 SIG5 rs7412 APOE 58.74 1.84e-59
22 SIG5 rs11887534 ABCG8 54.90 1.27e-55
23 SIG5 rs12740374 CELSR2 41.78 1.66e-42
24 SIG5 rs138294113 LDLR 38.23 5.90e-39
25 SIG5 rs9349379 PHACTR1 29.19 6.50e-30
26 SIG5 rs11591147 PCSK9 24.12 7.50e-25
27 SIG5 rs4977575 CDKN2B-AS1 19.93 1.18e-20
28 SIG5 rs2351524 NBEAL1 19.64 2.30e-20
29 SIG7 rs9275218 HLA-DQB1 16.26 5.53e-17
30 SIG7 rs9272451 HLA-DQA1 14.31 4.94e-15
31 SIG7 rs3184504 SH2B3 12.95 1.13e-13
32 SIG7 rs10774625 ATXN2 11.85 1.40e-12
33 SIG7 rs1275977 KCNK3 11.51 3.06e-12
34 SIG7 rs7192155 CFDP1 10.34 4.60e-11
35 SIG7 rs12509595 FGF5 9.80 1.57e-10
36 SIG7 rs2071278 NOTCH4 9.48 3.29e-10
37 SIG7 rs3806155 BTNL2 9.48 3.33e-10
38 SIG7 rs72831343 C10orf107 9.42 3.82e-10
In [4]:
# Load the presence matrix to identify unique Signature 5 loci
# Try multiple possible locations
present_matrix_paths = [
    Path("/Users/sarahurbut/Downloads/present_matrix_1mb_sig5.csv"),
    Path("/Users/sarahurbut/Library/CloudStorage/Dropbox-Personal/present_matrix_1mb_sig5.csv"),
]

present_matrix_file = None
for path in present_matrix_paths:
    if path.exists():
        present_matrix_file = path
        break

if present_matrix_file is not None and "loci_df" in locals():
    present_df = pd.read_csv(present_matrix_file)
    
    # The 10 unique Signature 5 SNPs (not in any constituent trait GWAS within 1MB)
    unique_sig5_snps = {
        'rs6687726': 'IL6R',
        'rs2509121': 'HYOU1',
        'rs4760278': 'R3HDM2',
        'rs1532085': 'LIPC',
        'rs7168222': 'NR2F2-AS1',
        'rs35039495': 'PLCG2',
        'rs8121509': 'OPRL1',
        'rs1499813': 'FNDC3B',
        '4:96088139': 'UNC5C',
        'rs4732365': 'C7orf55'
    }
    
    # Biological roles for each gene
    biological_roles = {
        'IL6R': 'Inflammation (Interleukin-6 receptor)',
        'HYOU1': 'Hypoxia/ER stress response',
        'R3HDM2': 'RNA binding protein',
        'LIPC': 'HDL metabolism (Hepatic lipase)',
        'NR2F2-AS1': 'Nuclear receptor antisense RNA',
        'PLCG2': 'Platelet activation, immunity',
        'OPRL1': 'Opioid receptor, stress/pain signaling',
        'FNDC3B': 'Insulin signaling, adipogenesis',
        'UNC5C': 'Axon guidance (Netrin receptor)',
        'C7orf55': 'Unknown function'
    }
    
    # Get Signature 5 loci
    sig5_loci = loci_df[loci_df['locus_SIG5'] == 1].copy()
    
    # Find the unique loci in our data
    unique_loci_results = []
    for rsid, expected_gene in unique_sig5_snps.items():
        if rsid.startswith('rs'):
            match = sig5_loci[sig5_loci['rsid'] == rsid]
        else:
            # For position-based IDs like 4:96088139
            match = sig5_loci[sig5_loci['UID'].str.contains(rsid.replace(':', ':'), na=False)]
        
        if len(match) > 0:
            row = match.iloc[0]
            unique_loci_results.append({
                'Rank': len(unique_loci_results) + 1,
                'rsID': row['rsid'],
                'Gene': row['nearestgene'],
                'Chr': row['#CHR'],
                'Position': row['POS'],
                'LOG10P': round(row['LOG10P'], 2),
                'P_value': f"{10**(-row['LOG10P']):.2e}",
                'Beta': round(row['BETA'], 4),
                'EAF': round(row['EAF'], 3),
                'Biological_Role': biological_roles.get(expected_gene, 'Unknown')
            })
        else:
            # SNP not found in our data - still include it
            unique_loci_results.append({
                'Rank': len(unique_loci_results) + 1,
                'rsID': rsid,
                'Gene': expected_gene,
                'Chr': '?',
                'Position': '?',
                'LOG10P': '?',
                'P_value': '?',
                'Beta': '?',
                'EAF': '?',
                'Biological_Role': biological_roles.get(expected_gene, 'Unknown')
            })
    
    unique_sig5_df = pd.DataFrame(unique_loci_results)
    
    print("="*80)
    print("THE 10 UNIQUE SIGNATURE 5 DISCOVERIES")
    print("="*80)
    print("Loci genome-wide significant in Signature 5 but NOT in constituent trait GWAS")
    print("(Angina, MI, Hypercholesterolemia, Coronary atherosclerosis, Acute IHD, Chronic IHD)")
    print("-"*80)
    print(f"\nFound {len([r for r in unique_loci_results if r['LOG10P'] != '?'])} of 10 unique loci in the data\n")
    
    display(unique_sig5_df[['Rank', 'rsID', 'Gene', 'LOG10P', 'P_value', 'Biological_Role']])
    
    print("\n" + "="*80)
    print("INTERPRETATION")
    print("="*80)
    print("These 10 loci represent novel genetic discoveries enabled by joint modeling.")
    print("Each locus has distributed pleiotropic effects across multiple cardiovascular")
    print("traits that are too weak to detect individually but collectively reach")
    print("genome-wide significance when analyzed jointly through Signature 5.")
    print("="*80)
else:
    if present_matrix_file is None:
        print("⚠️  Presence matrix file not found in any of these locations:")
        for path in present_matrix_paths:
            print(f"   {path}")
    if "loci_df" not in locals():
        print("⚠️  loci_df not found. Please run the previous cell to load the GWAS loci data.")

================================================================================
THE 10 UNIQUE SIGNATURE 5 DISCOVERIES
================================================================================
Loci genome-wide significant in Signature 5 but NOT in constituent trait GWAS
(Angina, MI, Hypercholesterolemia, Coronary atherosclerosis, Acute IHD, Chronic IHD)
--------------------------------------------------------------------------------

Found 10 of 10 unique loci in the data
Rank rsID Gene LOG10P P_value Biological_Role
0 1 rs6687726 IL6R 7.90 1.27e-08 Inflammation (Interleukin-6 receptor)
1 2 rs2509121 HYOU1 8.99 1.03e-09 Hypoxia/ER stress response
2 3 rs4760278 R3HDM2 7.50 3.17e-08 RNA binding protein
3 4 rs1532085 ALDH1A2 7.42 3.76e-08 HDL metabolism (Hepatic lipase)
4 5 rs7168222 NR2F2-AS1 7.58 2.66e-08 Nuclear receptor antisense RNA
5 6 rs35039495 PLCG2 7.58 2.63e-08 Platelet activation, immunity
6 7 rs8121509 OPRL1 8.43 3.74e-09 Opioid receptor, stress/pain signaling
7 8 rs1499813 FNDC3B 9.94 1.14e-10 Insulin signaling, adipogenesis
8 9 4:96088139_ATATG_A UNC5C 7.43 3.70e-08 Axon guidance (Netrin receptor)
9 10 rs4732365 C7orf55 8.94 1.14e-09 Unknown function 
================================================================================
INTERPRETATION
================================================================================
These 10 loci represent novel genetic discoveries enabled by joint modeling.
Each locus has distributed pleiotropic effects across multiple cardiovascular
traits that are too weak to detect individually but collectively reach
genome-wide significance when analyzed jointly through Signature 5.
================================================================================

4. Summary and Response¶

Key Findings¶

  1. Genome-wide significant loci identified: Multiple genetic loci are associated with signature exposure (151 total loci across 16 signatures).
  2. Signature-specific loci: Genetic variants associated with signatures but not with individual diseases.
  3. Novel discoveries: 10 unique loci for Signature 5 that are not found in constituent trait GWAS.
  4. Biologically plausible gene associations: Signature 5 is enriched for lipid metabolism genes (e.g., LDLR, APOE, PCSK9, LPA).

Response to Reviewer¶

We demonstrate biological meaningfulness through genetic association analysis. We performed GWAS using average signature exposure (AEX) as quantitative phenotypes, identifying genetic variants associated with disease signatures. Signature 5 (cardiovascular) is enriched for genes with known roles in lipid metabolism (e.g., LDLR, APOB, PCSK9, LPA), providing strong biological validation.

Critically, we identified 10 novel loci for Signature 5 that are genome-wide significant in the joint analysis but not detected in any individual constituent trait GWAS. This demonstrates that signature-based GWAS can discover genetic associations with distributed pleiotropic effects that are too weak to detect in single-disease analyses, providing direct evidence for the biological meaningfulness of our disease signatures.