Supplementary datasets for: Polymorphic short tandem repeats make widespread contributions to blood and serum traits

https://doi.org/10.5061/dryad.z612jm6jk

Please see the manuscript for method details of how these datasets were produced. Alternatively, the relevant portion of the methods from the paper have been pasted into the methods section on this page, with some distortions due to the difficulty of copying mathematical formulae.

The two supplementary datasets are:

STR association tests, by population and phenotype

These files are named {population}_{phenotype}_str_gwas_results.tab.gz Association statistics referred to in the paper are from the White British population unless otherwise specified. Association tests in the White British population were performed genome-wide, while STR association tests in the other populations were performed in fine-mapping regions identified by the White British association tests.

Populations are:

• white_british
• black
• south_asian
• chinese
• irish
• white_other

Phenotypes are written exactly as listed in Supplementary Table 1.
We filtered STRs with total minor allele dosage less than 20. This amounted to very few STRs for each population, phenotype combination. A list of those STRs (if any) are available per population and phenotype in the files named {population}_{phenotype}_str_gwas_results.filtered.tab

Association file columns:

• chromosome
• base_pair_location: beginning of the repeat (hg19, 1-indexed, inclusive)
• alleles: lengths alleles in the population measured in number of repeat units. For example, the allele 5 for an AC repeat implies the bases “ACACACACAC” (possibly with some impurity). Occasionally the repeat unit will be listed as none. This occurs when it was hard to determine the repeat unit from the period as there were multiple repeat units present in the reference allele of length equal to period and with similar frequencies. In that case, the period of the repeat will still be given, and the length of an allele in base pairs can still be calculated by multiplying the allele by the period.
• beta: measured effect size of the linear association of the rank-inverse-normalized phenotype against the length-dosages of unnormalized STR genotypes, measured in number of repeat units. Phenotypes are measured in unspecified units as they are rank-inverse-normalized, so these betas should only be compared to betas from other studies with sufficient reason to believe that such a comparison is meaningful. p-values may be more comparable between studies.
• standard_error: See caveats for beta
• allele_frequencies
• p_value: p-values less than 1e-300 exceeded our software’s numeric precision and are listed as 0
• ref_allele: measured in number of repeat units
• repeat_unit: the standardized repeat unit of this STR, or none if there was no one clear repeat unit
• period: the length of the repeat unit
• end_pos (hg19): end of the repeat (1-indexed, inclusive)
• start_pos (hg38)
• end_pos (hg38)
• n: this study worked with imputed calls and no call-level filters, as such n will be equivalent for each variant associated with the same phenotype
• number_of_common_alleles: the number of alleles in the population with frequency >= 1%
• mean_{phenotype}_per_summed_gt: the mean phenotype value for each sum of allele lengths, where each participant’s contribution to the phenotype mean for each length-sum is weighted by the imputed probability of their true genotype sum being equal to that length-sum. Can be used for plotting graphs of mean phenotype value vs summed-length. Summed gts are measured in number of summed repeat units.
• summed_0.05_significance_CI: The 95% symmetric confidence interval for each of the means above
• summed_5e-8_significance_CI: The (1 - 5e-8) symmetric confidence interval for each of the means above. I.e. this interval is expected to contain the true mean with a probability of 1 - 5e-8, which is very close to one.
• mean_{phenotype}_per_paired_gt: the mean phenotype value for each unordered pair of allele lengths, where each participant’s contribution to the phenotype mean for each pair is weighted by the imputed probability of their true genotype pair being equal to that pair. Can be used for plotting graphs of mean phenotype value vs length pair. Each gt in each pair is measured in number of repeat units.
• paired_0.05_significance_CI: The 95% symmetric confidence interval for each of the means above
• paired_5e-8_significance_CI: The (1 - 5e-8) symmetric confidence interval for each of the means above

In addition, the filtered files have the column locus_filtered which has the same value MAC<20 for each row.

Tarballs of fine-mapping outputs by phenotype

Fine-mapping tarballs are named finemapping_{phenotype}.tgz. Each tarball contains two summary tables: finemapping_first_pass.tab and finemapping_followup.tab, describing the results from standard fine-mapping runs and followup fine-mapping runs under alternate conditions, respectively. Additionally, each tarball contains all the raw fine-mapping outputs for each fine-mapping run in each trait-region for both SuSiE and FINEMAP. Note that fine-mapping was only performed in the White British population.

Summary file (finemapping_first_pass.tab and finemapping_followup.tab) notes and column descriptions:

• positions are hg19 and 1-indexed
• Regions are denoted by {chrom}_{start_pos_inclusive}_{end_pos_inclusive}
• variant names for SNPS are SNP_{position}_{ref}_{alt}
• variant names for STRs are STR_{start_position}
• p_val, coeff and se are from the association of that variant with the trait being fine-mapped
• susie_cs values denote which pure cs a variant is present in, or -1 if its in no pure cs
• finemap_pip and susie_alpha values were used as CP values in the paper
• susie_cs values of -1 imply that the susie_alpha value should be ignored (treated as zero)
• only variants included in fine-mapping are present in these files
• some variants lack susie values if they were not included in the susie fine-mapping (probably because they had too high a p-value) The followup files only contain variants from the regions subjected to follow-up fine-mapping conditions.
• traits with no regions that were followed-up on are empty.
• In addition to the columns in the first pass file (except for coeff or se), there are additional FINEMAP or SuSiE columns for each extra condition.
• the best_guess column corresponds to the use of best guess genotypes from imputation for fine-mapping
• the ratio columns correspond to the prior of favoring SNP over STR causality by a 4-to-1 ratio
• the repeat column corresponds to the repeat FINEMAP run with no changed settings
• the total_prob column corresponds to the FINEMAP run with a prior of there being 4 total causal variants
• the prior_std_derived and prior_std_low columns correspond to the priors for the effect sizes of causal variants of 0.05% and 0.0025%, respectively
• the conv_tol column corresponds to the flag –prob-conv-sss-tol 0.0001
• the mac column corresponds to the non-major allele dosage threshold of 100
• the p_thresh column corresponds to the p_value threshold of 1e-4
• values for the additional columns will be missing for variants which were not fine-mapped in specifically those conditions (say, a variant with p-value of 1e-3 in the 1e-4 threshold column)

Raw fine-mapping output file descriptions

This consists of files for both FINEMAP and SuSiE.
The files for FINEMAP are ( described at http://christianbenner.com/ , or screenshots of the website pasted below if the URL doesn't work):

FINEMAP_first_pass_{region}_finemap_output.log
FINEMAP_first_pass_{region}_finemap_output.snp
FINEMAP_first_pass_{region}_finemap_output.config
FINEMAP_first_pass_{region}_finemap_output.credX

and the following files for SuSiE:

SuSiE_first_pass_{region}_alpha.tab
SuSiE_first_pass_{region}_colnames.txt
SuSiE_first_pass_{region}_csX.txt
SuSiE_first_pass_{region}_lbf.tab
SuSiE_first_pass_{region}_lbf_variable.tab
SuSiE_first_pass_{region}_lfsr.tab
SuSiE_first_pass_{region}_sigma2.txt
SuSiE_first_pass_{region}_V.tab

• Except for the colnames and cs files, these are arrays written from the output fields of the susie() function described at https://stephenslab.github.io/susieR/reference/susie.html (with screenshots pasted below if the URL doesn't work)
• colnames contains one variant name per line, each line corresponding to one column of the alpha array
• csX contains three rows:
  • the first contains the 1-indexed numbers of the variables included in the credible set, in ascending order
  • the second contains the coverage of the credible set (Always greater than 0.9 which was the requested coverage)
  • the third contains three numbers, the min, mean and median absolute correlations between each variable in the credible set (or 100 randomly subsampled variables if there are more than 100 variables)
  • the 1-indexed number in the filename corresponds to the corresponding column of the alpha.tab array

Additionally, for each of the regions we followed up on, this tarball contains the same files as above, but with the following prefixes, corresponding to the follow-up fine-mapping condition being tested:

FINEMAP_derived_effect_size_prior - (effect size prior of 0.05%)
FINEMAP_low_effect_size_prior - (effect size prior of 0.0025%)
FINEMAP_mac_threshold_100 - (non-major allele dosage threshold of 100)
FINEMAP_prior_4_signals - (prior of 4 causal variants per region)
FINEMAP_prior_snps_over_strs - (prior of favoring SNP over STR causality by a 4-to-1 ratio)
FINEMAP_pval_threshold_1e4 - (p_value threshold of 1e-4)
FINEMAP_stricter_stopping_threshold - (using the flag --prob-conv-sss-tol 0.0001)
SuSiE_prior_snps_over_strs - (prior of favoring SNP over STR causality by a 4-to-1 ratio)
SuSiE_best_guess_genotypes - (the use of best guess genotypes from imputation for fine-mapping)

See the paper and Supplementary Note 3 for more details on each follow-up condition.
We do not have raw output files for FINEMAP_repeat runs.
Notes:

• For FINEMAP, I have focused on the posterior probabilities in the finemap_output.snp files
• For SuSiE I focused on the cs files with min absolute correlation > 0.8, and then the values in alpha.tab with rows identified by the CS number and columns identified by the variants in the first row of the CS file
• the log files for FINEMAP saying v1.4.1 seem to be buggy, the actual FINEMAP output to the command line indicates that v1.4.2 was run.

Sharing/Access information

This data is also accessible at https://gymreklab.com/science/2023/09/08/Margoliash-et-al-paper.html

Code/Software

The code repository used to produce these results is https://github.com/LiterallyUniqueLogin/ukbiobank_strs and is available as a repository frozen at the time of publication on Zenodo at the DOI: https://doi.org/10.5281/zenodo.8436632

FINEMAP documentation screenshots

SuSiE documentation screenshots