# Per-phenotype files

##### note

Please note that p-values are now stored as natural log p-values to avoid underflow (i.e., ln P, not -ln P or -log10 P).

## Overview#

The data are released in 7,228 flat files, one for each phenotype, and a corresponding tabix index file for each. These files are available on Amazon AWS (for large-scale analysis, we recommend using the Hail format files on Google Cloud).

The files are named with respect to their trait_type, phenocode, and a combination of pheno_sex, coding, or modifier. To find a specific phenotype, we suggest looking in the phenotype manifest (Google Sheets) (available for download on Amazon). Search for your phenotype(s) of interest and use the paths indicated to download the summary statistics. A description of fields in the manifest can be found here.

The per-phenotype files are summary statistics files containing meta-analyzed and single-ancestry GWAS results. We especially highlight the low_confidence fields, which includes some (non-exhaustive) basic quality control filters (see below). These files each have 28,987,534 variants, but note that not all populations will have data for each variant.

Finally, the variant manifest file includes information on each variant in the dataset and has the same number of rows as each per-phenotype file. We highlight the high_quality column which represents variants that are PASS variants in gnomAD and have consistent frequencies with each population in gnomAD (AFR, AMR, EAS, and EUR frequencies are within 2-fold or chi-squared p-value of the difference > 1e-6).

## Phenotype manifest file#

### Phenotype ID fields#

The first 5 fields are guaranteed to be unique.

trait_type: One of the following: continuous, biomarkers, prescriptions, icd10, phecode, categorical

phenocode: The code for the phenotype (for continuous, biomarkers, and categorical traits, this corresponds to the field ID as described by UKB, e.g. 21001 for BMI)

pheno_sex: Indicating whether the phenotype was run for both sexes (pheno_sex="both_sexes") or in just females (pheno_sex="females") or males (pheno_sex="males"). In 0.1, this is only differentiated for phecodes.

coding: For categorical variables, this corresponds to the coding that was used (e.g. coding 2 for field 1747). For all other trait_types, this field is blank.

modifier: Refers to any miscellaneous downstream modifications of the phenotype (e.g. irnt for inverse-rank normal transformation). If the phenotype is updated, this field can be used to denote the update (e.g. the particular wave of COVID-19 data used).

description: A shorter description of the phenotype (for continuous, biomarkers, and categorical variables, corresponds to the Description on the showcase). For phecodes, this is the "description" column in the phecodes definition file.

description_more: A longer description of the phenotype (for continuous and categorical variables, corresponds to the Notes page on the showcase).

coding_description: For categorical variables, a description of the particular coding that was used (the Meaning column on the showcase page for that coding).

category: A categorization of the phenotype. For continuous, biomarkers, and categorical traits, this corresponds to the Category at the top of the showcase page. For ICD codes, this corresponds to the Chapter of the ICD code; for phecodes, this is the "group" column in the phecodes definition file; for prescriptions, this corresponds to a semi-manual categorization of prescription drugs.

in_max_independent_set: If the phenotype is in our maximally indepdent set. This set of relatively uncorrelated phenotypes was constructed using a pairwise phenotypic correlation matrix of phenotypes with ancestries passing all QC filters (released here via make_pairwise_ht). Of all phenotype pairs, we retained any with a pairwise correlation $r < 0.1$. For pairs with $r > 0.1$ , we used hl.maximal_independent_set to identify indendent phenotypes for retention, imposing a tiebreaker of higher case count (or higher sample size for continuous phenotypes), producing 195 independent phenotypes.

### Case and ancestry fields#

##### note

If a trait is quantitative (trait_type is "continuous" or "biomarkers"), all samples are considered to be "cases". Thus, the number of cases is equivalent to the number of samples.

n_cases_full_cohort_both_sexes: Number of cases (or individuals phenotyped for quantitative traits) across all ancestry groups, females and males combined. Should be similar to the sum of per-ancestry n_cases for relevant ancestries, but may include ancestry outliers and samples that failed QC.

n_cases_full_cohort_females: Number of female cases (or individuals phenotyped for quantitative traits) across all ancestry groups. May include ancestry outliers and samples that failed QC.

n_cases_full_cohort_males: Number of male cases (or individuals phenotyped for quantitative traits) across all ancestry groups. May include ancestry outliers and samples that failed QC.

n_cases_hq_cohort_both_sexes : Number of cases (or individuals phenotyped for quantitative traits) across ancestry groups passing stringent phenotype QC (see pops_pass_qc), females and males combined. Should be similar to the sum of per-ancestry n_cases for relevant ancestries, but may include ancestry outliers and samples that failed QC.

n_cases_hq_cohort_females: Number of female cases (or individuals phenotyped for quantitative traits) across ancestry groups passing stringent phenotype QC (see pops_pass_qc). May include ancestry outliers and samples that failed QC.

n_cases_hq_cohort_males: Number of male cases (or individuals phenotyped for quantitative traits) across ancestry groups passing stringent phenotype QC (see pops_pass_qc). May include ancestry outliers and samples that failed QC.

pops: Comma-delimited list of ancestry codes for which this phenotypes was GWASed.

num_pops: Number of ancestry groups for which this phenotype was GWASed.

pops_pass_qc: Comma-delimited list of ancestry codes for which this phenotype passes QC (see quality control, heritability manifest, and phenotype_qc_{pop} field).

num_pops_pass_qc: Number of ancestry groups for which this phenotype passes QC.

### Population-specific fields#

##### note

The variable pop is a placeholder for a 3-letter ancestry code. For example, n_cases_AFR is the number of cases with AFR ancestry.

##### note

If a trait is quantitative (trait_type is "continuous" or "biomarkers"), all samples are considered to be "cases". Thus, the number of cases is equivalent to the number of samples.

##### note

The final heritability estimates for non-EUR traits use a randomized Haseman Elston estimator on genotype data with 5 MAF and 5 LD bins and 50 random variables (rhemc_25bin_50rv), while for computational tractability for EUR we use S-LDSC with 5 MAF and 5 LD bins. In the phenotype manifest we provide estimates with the sldsc_25bin_ prefix for EUR and the rhemc_25bin_50rv_ prefix for other ancestry groups. See the heritability manifest for all heritability estimates computed, including sldsc_25bin for all ancestry groups.

n_cases_{pop}: Number of cases (or individuals phenotyped for quantitative traits) with pop ancestry in the GWAS analysis. Excludes ancestry outliers and samples that failed QC.

n_controls_{pop}: Number of controls with pop ancestry in the GWAS analysis. Excludes ancestry outliers and samples that failed QC.

{rhemc_25bin_50rv/sldsc_25bin}_h2_observed_{pop}: Observed scale heritability estimates using 25 MAF/LD bins with RHEmc (non-EUR) or SLDSC (EUR).

{rhemc_25bin_50rv/sldsc_25bin}_h2_observed_se_{pop}: Observed scale heritability standard error using 25 MAF/LD bins with RHEmc (non-EUR) or SLDSC (EUR).

{rhemc_25bin_50rv/sldsc_25bin}_h2_liability_{pop}: Libaility scale heritability estimates using 25 MAF/LD bins with RHEmc (non-EUR) or SLDSC (EUR), transformed using per-ancestry in-sample prevalence.

{rhemc_25bin_50rv/sldsc_25bin}_h2_liability_se_{pop}: Liability scale heritability standard error using 25 MAF/LD bins with RHEmc (non-EUR) or SLDSC (EUR), transformed using per-ancestry in-sample prevalence.

{rhemc_25bin_50rv/sldsc_25bin}_h2_z_{pop}: Heritability Z-scores (for test of h2 > 0) using per-ancestry-trait pair h2 estimates and standard errors.

lambda_gc_{pop}: The genomic control (lambda GC) calculated from the summary statistics for pop with low-confidence statistics removed and only considering high-quality variants.

phenotype_qc_{pop}: Phenotype QC outcome for each ancestry-trait pair. Filters are described in the heritability manifest in more detail. Filters are applied sequentially; this field specifies either PASS or the reason for failure.

### File information#

##### note

For each field in this section there also exists a field with the suffix _tabix, which contains the equivalent information for the tabix file. For instance, filename_tabix contains the name of the tabix file.

filename: Name of summary statistics file.

aws_link: Link to download summary statistics file from Amazon AWS.

## Heritability manifest file#

### Phenotype ID fields#

##### note

The variable pop is a placeholder for a 3-letter ancestry code. For example, n_cases_AFR is the number of cases with AFR ancestry.

##### note

Unlike the phenotype manifest, the heritability manifest provides a row for each ancestry group. Thus the first 6 fields are guarenteed to be unique.

trait_type: One of the following: continuous, biomarkers, prescriptions, icd10, phecode, categorical

phenocode: The code for the phenotype (for continuous, biomarkers, and categorical traits, this corresponds to the field ID as described by UKB, e.g. 21001 for BMI)

pheno_sex: Indicating whether the phenotype was run for both sexes (pheno_sex="both_sexes") or in just females (pheno_sex="females") or males (pheno_sex="males"). In 0.1, this is only differentiated for phecodes.

coding: For categorical variables, this corresponds to the coding that was used (e.g. coding 2 for field 1747). For all other trait_types, this field is blank.

modifier: Refers to any miscellaneous downstream modifications of the phenotype (e.g. irnt for inverse-rank normal transformation). If the phenotype is updated, this field can be used to denote the update (e.g. the particular wave of COVID-19 data used).

heritability.pop: Ancestry group.

### Heritability methods#

##### note

Each of the below Heritability methods subheadings contain (a subset of) the fields described in the Heritability estimates section (except for SAIGE estimates, which are provided as-is).

##### note

rhemc_8bin was run only for a subset of traits in the EUR ancestry group and for all traits across non-EUR ancestry groups. rhemc_25bin and rhemc_25bin_50rv were run only for non-EUR ancestry groups due to computational complexity. sldsc_25bin and ldsc were run for all ancestry-trait pairs.

##### note

More information can be found here on the heritability estimation approach, and important caveats can be found here.

heritability.estimates.ldsc.*: Univariate LD score regression

heritability.estimates.sldsc_25bin.*: Stratified LD score regression, 5 LD score bins x 5 MAF bins

heritability.estimates.rhemc_25bin.*: RHEmc (HE regression), 5 LD score bins x 5 MAF bins

heritability.estimates.rhemc_8bin.*: RHEmc (HE regression), 4 LD score bins x 2 MAF bins

heritability.estimates.rhemc_25bin_50rv.*: RHEmc (HE regression), 5 LD score bins x 5 MAF bins; 50 random variables for improved power

heritability.estimates.saige: The heritability as estimated by SAIGE: note that this is likely not well-calibrated for binary traits, or traits with high heritabilities.

heritability.estimates.final.*: Final estimates; 25 bin SLDSC for EUR and 25 bin, 50 RV RHEmc for non-EUR (these are also present in the full manifest)

### Heritability estimates#

heritability.estimates.*.h2_observed: Observed scale heritability point estimates.

heritability.estimates.*.h2_liability: Liability scale results, using per-ancestry in-sample prevalence.

heritability.estimates.*.h2_z: Heritability Z-scores for test of $h^2 > 0$.

heritability.estimates.*.intercept: LDSC intercept. Only present for ldsc and sldsc_25bin

heritability.estimates.*.ratio: LDSC ratio, given by $\frac{intercept-1}{mean \chi^2 -1}$ as a measure of the proportion of test statistic inflation not attributed to polygenicity. Only present for ldsc and sldsc_25bin.

### QC flags#

##### note

The below QC flags were applied sequentially. See quality control for more information.

heritability.N_ancestry_QC_pass: Number of ancestries passing all QC per trait.

heritability.qcflags.GWAS_run: if the GWAS was performed for the ancestry-trait pair

heritability.qcflags.ancestry_reasonable_n: if the ancestry has a reasonable sample size; has the effect of removing AMR

heritability.qcflags.defined_h2: if the heritability estimate is non-missing

heritability.qcflags.significant_z: if the ancestry-trait pair shows $h^2$ z-score $> 0$

heritability.qcflags.in_bounds_h2: if, for all ancestries for a given trait, observed-scale heritability estimates $\in (0,1)$

heritability.qcflags.normal_lambda: if, for all ancestries for a given trait, $\lambda_{GC} > 0.9$

heritability.qcflags.normal_ratio: if, for the top three best powered ancestry groups (EUR, CSA, AFR), the S-LDSC ratio, given by $\frac{intercept-1}{mean \chi^2 -1}$, $< 0.3$ or the ratio z-score $< 4$

heritability.qcflags.EUR_plus_1: if the trait passes all above filters in EUR and at least 1 other ancestry group

heritability.qcflags.pass_all: if all QC flags pass for each ancestry-trait pair

## Per-phenotype files#

##### note

All p-values reported in the summary statistics flat files are now stored as log p-values to avoid underflow. This includes the per-ancestry association test p-value as well as pval_meta, pval_meta_hq, pval_heterogeneity, and pval_heterogeneity_hq.

The per-phenotype files are tsv.bgz files are (b)gzipped: they can either be unzipped (zcat file.tsv.bgz > file.txt), or read natively in R (read_delim(gzfile('file.tsv.bgz'), delim='\t')) and Python (gzip.open('file.tsv.bgz')).

Depending on whether a phenotype is quantitative (trait_type is "continuous" or "biomarkers") or binary (trait_type is "prescriptions", "icd10", "phecode" or "categorical"), the number of columns will change due to case/control-stratified statistics for binary phenotypes.

### Variant fields#

chr: Chromosome of the variant.

pos: Position of the variant in GRCh37 coordinates.

ref: Reference allele on the forward strand.

alt: Alternate allele (not necessarily minor allele). Used as effect allele for GWAS.

### Meta-analysis fields#

##### note

All meta-analyses were only performed on variants that were not flagged as low_confidence for a given population. As described below in Population-specific fields, per-variant low_confidence status is specific to each ancestry-trait pair.

af_meta: Alternate allele frequency from meta-analysis across populations for which this phenotype was GWASed. NOTE: This field only appears in files for quantitative phenotypes.

af_cases_meta: Alternate allele frequency in cases from meta-analysis across populations for which this phenotype was GWASed. NOTE: This field only appears in files for binary phenotypes.

af_controls_meta: Alternate allele frequency in controls from meta-analysis across populations for which this phenotype was GWASed. NOTE: This field only appears in files for binary phenotypes.

beta_meta: Estimated effect size of alternate allele from meta-analysis across populations for which this phenotype was GWASed.

se_meta: Estimated standard error of beta_meta.

pval_meta: log p-value of beta_meta significance test.

pval_heterogeneity: log p-value from heterogeneity test of meta-analysis.

### High quality meta-analysis fields#

##### note

These fields are only present in flat files for traits that have any QC-pass ancestries. As a requirement for passing QC a trait must pass in EUR and at least 1 other ancestry (see quality control).

##### note

As above, meta-analyses were only performed on variants that were not flagged as low_confidence for a given population. As described below in Population-specific fields, per-variant low_confidence status is specific to each ancestry-trait pair.

af_meta_hq: Alternate allele frequency from meta-analysis across populations for which this phenotype passes all QC filters. NOTE: This field only appears in files for quantitative phenotypes.

af_cases_meta_hq: Alternate allele frequency in cases from meta-analysis across populations for which this phenotype passes all QC filters. NOTE: This field only appears in files for binary phenotypes.

af_controls_meta_hq: Alternate allele frequency in controls from meta-analysis across populations for which this phenotype passes all QC filters. NOTE: This field only appears in files for binary phenotypes.

beta_meta_hq: Estimated effect size of alternate allele from meta-analysis across populations for which this phenotype passes all QC filters.

se_meta_hq: Estimated standard error of beta_meta_hq.

pval_meta_hq: log p-value of beta_meta_hq significance test.

pval_heterogeneity_hq: log p-value from heterogeneity test of meta-analysis.

### Population-specific fields#

##### note

The variable pop used in this section is a placeholder for a 3-letter ancestry code. For example, af_AFR is the alternate allele frequency for AFR samples included in the GWAS of this phenotype.

##### note

An ancestry-specific column is only included in the file if a GWAS was run for that ancestry. For example, a trait that was only GWASed in AMR and CSA samples will only have the fields af_AMR, af_CSA, beta_AMR, beta_CSA, etc.

af_{pop}: Alternate allele frequency for pop samples included in the GWAS of this phenotype. NOTE: This field only appears in files for quantitative phenotypes.

af_cases_{pop}: Alternate allele frequency for pop cases included in the GWAS of this phenotype. NOTE: This field only appears in files for binary phenotypes.

af_controls_{pop}: Alternate allele frequency for pop controls included in the GWAS of this phenotype. NOTE: This field only appears in files for binary phenotypes.

beta_{pop}: Estimated effect size of alternate allele from GWAS of pop samples.

se_{pop}: Estimated standard error of beta_{pop}.

pval_{pop}: log p-value of beta_{pop} significance test.

low_confidence_{pop}: Boolean flag indicating low confidence for pop based on the following heuristics:

• Alternate allele count in cases <= 3
• Alternate allele count in controls <= 3
• Minor allele count (cases and controls combined) <= 20

## Variant manifest file#

### Variant fields#

As in per-phenotype files.

chr: Chromosome of the variant.

pos: Position of the variant in GRCh37 coordinates.

ref: Reference allele on the forward strand.

alt: Alternate allele (not necessarily minor allele). Used as effect allele for GWAS.

rsid: The RSID for the variant (from the BGEN file from UK Biobank).

varid: The variant ID for the variant (from the BGEN file from UK Biobank).

pass_gnomad_genomes: A boolean corresponding to the PASS status in gnomAD (NA if variant is not in gnomAD).

n_passing_populations: The number of populations (max 4: AFR, AMR, EAS, EUR) where the frequency in UKB is less than twice the frequency in gnomAD for the corresponding population (see below), and the p-value of a chi-squared test assessing the difference is > 1e-6.

high_quality: A boolean corresponding to a high-quality variant based on these filters (pass_gnomad_genomes & n_passing_populations == 4).

nearest_genes: The nearest genes for this variant based on Gencode v19.

info: The Info score for this variant (from the ukb_mfi_chrN_v2.txt files).

### Population-specific fields#

##### note

The variable pop used in this section is a placeholder for a 3-letter ancestry code. For example, af_AFR is the alternate allele frequency for AFR individuals in the whole dataset.

ac_{pop}: The alternate allele count for this variant across all individuals in pop. Defined as af_{pop} * an_{pop}.

af_{pop}: The alternate allele frequency for this variant across all individuals in pop. The mean dosage divided by two.

an_{pop}: The alternate allele number for this variant. This is twice the number of pop individuals with a defined genotype at this site.

gnomad_genomes_ac_{pop}: The alternate allele count for this variant in the nearest gnomAD population: AFR, EAS, and AMR are matched as-is, while EUR is matched to the "North-West European" subset of gnomAD (not available for CSA or MID, as these populations are not in gnomAD v2 genomes).

gnomad_genomes_af_{pop}: The alternate allele frequency for this variant in the nearest gnomAD population.

gnomad_genomes_an_{pop}: The alternate allele number for this variant in the nearest gnomAD population. This is twice the number of individuals with a defined genotype at this site.