GAMBIT

A C++ tool for Gene-based Analysis with oMniBus, Integrative Tests

• Implements several gene-based test forms (quadratic: weighted sum of Zsq, linear: weighted sum of Z, and maximum Zsq) to aggregate GWAS single-variant summary statistics cross-referenced with variant- or region-based functional annotations
• Calculates annotation-stratified gene-based tests (e.g., TWAS/PrediXcan tests using eSNPs, gene-based tests using only coding variants, and gene-based tests using enhancer-to-target-gene maps), and omnibus tests by combining p-values for each gene
• Inputs: GWAS association summary statistics file (chromosome, position, ref/alt allele, and z-score or beta-hat + se), annotation files, and LD reference panel

GWAS Summary Statistics

• GWAS summary statistics files can be specified via --gwas my_summary_stats.txt.gz. Input files must be ordered by chromosome and genomic position, with input fields as shown below:

#CHR  POS     REF  ALT  SNP_ID      N         ZSCORE   ANNO
1     721290  C    G    rs12565286  58663.62  0.86661  Intergenic
1     752566  G    A    rs3094315   57135     0.5521   Intergenic
1     775659  A    G    rs2905035   54570     1.12098  Intron:LOC643837
1     777122  A    T    rs2980319   54570     1.11906  Exon:LOC643837

• The first four fields and ZSCORE are required, while SNP_ID, ANNO and N (effective sample size) are optional.
• See format_gwas_summary_stats.sh for annotating GWAS summary statistics files using EPACTS/TabAnno.

Annotation-Stratified Gene-Based Tests

Gene-Based Analysis with Regulatory Elements

• To compute gene-based tests using regulatory element annotations, specify an annotation bed file with regulatory-element-to-target-gene weights via --anno-bed my_reg_elems.txt.gz, formatted

#CHR  START   END     CLASS     ELEMENT_ID          TARGET_GENES                     ANNO
chr1  567400  567600  Enhancer  chr1:567400:567600  MIB2:4.12|CPTP:2.53|GLTPD1:2.53  .
chr1  568000  568200  Enhancer  chr1:568000:568200  ATAD3A:2.75                      .
chr1  758600  758800  Enhancer  chr1:758600:758800  C1orf170:2.57|PERM1:2.57         .
chr1  769200  769400  Enhancer  chr1:769200:769400  C1orf170:3.36|PERM1:3.36         .

• Association tests for individual regulatory elements is reported in *.stratified_out.txt files, and gene-based p-values (aggregating across regulatory elements for each gene) in *.summary_out.txt files.

• Aggregation Methods for Regulatory Elements. By default, GAMBIT aggregates test statistics across variants in regulatory elements using a weighted sum of single-variant chi-squared statistics (SKAT gene-based test). To instead use weighted ACAT or HMP to combine single-variant p-values, specify --no-skat and a p-value combination method via --pcomb.

Gene-Based Analysis with Coding and Other Annotated Variants

• To compute gene-based tests using coding and other variants, GAMBIT relies on the ANNO field in GWAS summary statistics and an annotation hierarchy definitions file specified via --anno-defs my_defs.txt, formatted as below:

#CLASS    SUBCLASS          ANNO_TERMS
Coding    Protein_Altering  Nonsynonymous,Start_Loss,Stop_Gain,Stop_Loss,CodonGain,CodonLoss,Frameshift
Coding    Splice_Site       Essential_Splice_Site,Normal_Splice_Site
Coding    Exon_Other        Exon,Synonymous
UTR       UTR3              Utr3
UTR       UTR5              Utr5

• The ANNO_TERMS field specifies a comma-separated list of annotation terms (matching terms from the GWAS summary statistics file's ANNO field), and CLASS and SUBCLASS determine the annotation hierarchy and classes reported in output files.

• Gene-Based Test Output. Test statistics stratified by gene and annotation subclass are provided in *.stratified_out.txt files, and gene-based p-values (aggregating across annotation classes for each gene) in *.summary_out.txt files.

• Variant Aggregation Methods. By default, GAMBIT aggregates test statistics across variants using a weighted sum of single-variant chi-squared statistics (SKAT gene-based test). To instead use weighted ACAT or HMP to combine single-variant p-values, specify --no-skat and a p-value combination method via --pcomb.

TWAS Analysis

• To compute TWAS/PrediXcan gene-based tests using GAMBIT, specify an eWeight file via --eweights my_eWeights.txt.gz, formatted

##TISSUE_IDS=0:Adipose_Subcutaneous,1:Adipose_Visceral_Omentum,2:Adrenal_Gland,3:Artery_Aorta
#CHR  POS     RSID       REF  ALT  BETAS
1     752566  rs3094315  G    A    C1orf159=3.92e-02@0|UBE2J2=-1.49e-02@0|FAM87B=2.75e-01@1;1.25e-01@2;1.17e-01@3
1     752721  rs3131972  A    G    LINC00115=1.15e-01@0;1.75e-02@3;4.90e-02@4|RP11-206L10.8=3.21e-02@1
1     754182  rs3131969  A    G    LINC00115=-2.1e-02@1|RP5-857K21.2=-8.27e-02@2|RP11-206L10.9=-1.11e-01@2
1     760912  rs1048488  C    T    C1orf159=3.35e-04@0|TTLL10=-1.4e-02@3|FAM87B=1.75e-01@1;1.12e-01@2;9.51e-02@3|SAMD11=-1.27e-02@2

• The BETAS field format is eGene_A=Weight_A1@Tissue_A1;Weight_A2@Tissue_A2|eGene_B=Weight_B1@Tissue_B1, and labels for tissue IDs can be specified in the header.
• Subsetting tissues. To restrict analysis to a subset of tissues/cell-types, specify a comma-separated list of tissues following the --tissues flag. By default, GAMBIT includes all tissues/cell-types present in the eWeight file.
• Tissue Aggregation for Omnibus tests. GAMBIT reports both single-tissue TWAS/PrediXcan analysis results, and omnibus tests results aggregating across all specified tissues/cell-types for each eGene. Omnibus p-values for multi-tissue TWAS/PrediXcan analysis can be calculated in GAMBIT using either 1) the maximum single-tissue test statistic based on the joint distribution of single-tissue statistics, 2) the sum of squared single-tissue z-scores (analogous to SKAT), or 3) PCOMB for ACAT or HMP [default]. Omnibus test method for multi-tissue analysis can be specified via --tissue-aggreg (PCOMB, MinP, SKAT, or ALL). P-value combination method can be specified via --pcomb (ACAT or HMP).
• Single-tissue and omnibus test output. Gene-based tests and p-values for each eGene-tissue pair are reported in *.stratified_out.txt files, and omnibus p-values (aggregating across all tissues for each eGene) in *.summary_out.txt files.

dTSS-Weighted Gene-Based Tests

• To incorporate un-annotated regulatory variants in gene-based analysis, GAMBIT implements a dTSS (distance to Transcription Start Site) weighted gene-based test, which aggregates all single-variant p-values within a specified window from each gene's TSS using weighted ACAT or HMP and assigns higher weight to variants nearer the TSS using an exponential decay function.
• To compute dTSS-weighted gene-based tests, specify a TSS bed file via --tss-bed my_tss_bed.bed.gz, fomatted

#CHR  START   END     SYMBOL      GENE             GENE_ANNO
1     11868   11869   DDX11L1     ENSG00000223972  transcribed_unprocessed_pseudogene
1     62947   62948   OR4G11P     ENSG00000240361  transcribed_unprocessed_pseudogene
1     69090   69091   OR4F5       ENSG00000186092  protein_coding
1     131024  131025  CICP27      ENSG00000233750  processed_pseudogene

• Window size. The window size for dTSS-weighted gene-based tests can be modified by specifying --tss-window BASEPAIRS (500 Kbp by default).
• dTSS decay function. The relative weight assigned to variants nearer/farther from the TSS can be modified by specifying --tss-alpha ALPHA, where alpha=0 implies all variants receive equal weight, and larger values confer more weight to variants nearer the TSS. --tss-alpha also accepts comma-separated lists of alpha values, in which case GAMBIT computes global test p-values across all specified values (individual p-values are reported in INFO output field). By default, GAMBIT uses dTSS alpha values 1e-4,5e-5,1e-5,5e-6.

Methods References

Statistical methods implemented in GAMBIT:

• Sequence Kernel Association Test (SKAT): Wu et al. (2011), AJHG
• TWAS: Gusev et al. (2016), Nat Genet
• PrediXcan: Gamazon et al. (2015), Nat Genet and Barbeira et al. (2018), Nat Comm
• Aggregated Cauchy Association Test (ACAT): Liu et al. (2019), AJHG and Liu and Xie (2018), arXiv
• Asymptotically exact Harmonic Mean P-value (HMP): Wilson (2019), PNAS

Software References

Libraries and resources used or adapted in GAMBIT:

• PDF, CDF, and quantile functions
  • ROOT system library, CERN
  • CDFLIB, Brown et al.
  • eigenmvn, Benazera et al.
  • libMvtnorm, Zhan et al.
• Hopscotch hashing
  • tsl library, Tessil et al.
• Tabix and HTSLIB
  • tabixpp, ekg et al.
  • htslib, samtools team
• Matrix libraries
  • Eigen matrix library

Feedback and bug reports

• Feel free to contact Corbin Quick (corbinq@gmail.com) with questions, bug reports, or feedback