format_sumstats: Check that summary statistics from GWAS are in a homogeneous...

In neurogenomics/MungeSumstats: Standardise summary statistics from GWAS

Check that summary statistics from GWAS are in a homogeneous format

Description

Check that summary statistics from GWAS are in a homogeneous format

Usage

format_sumstats(
  path,
  ref_genome = NULL,
  convert_ref_genome = NULL,
  chain_source = "ensembl",
  local_chain = NULL,
  convert_small_p = TRUE,
  convert_large_p = TRUE,
  convert_neg_p = TRUE,
  compute_z = FALSE,
  force_new_z = FALSE,
  compute_n = 0L,
  convert_n_int = TRUE,
  impute_beta = FALSE,
  es_is_beta = TRUE,
  impute_se = FALSE,
  analysis_trait = NULL,
  ignore_multi_trait = FALSE,
  INFO_filter = 0.9,
  FRQ_filter = 0,
  pos_se = TRUE,
  effect_columns_nonzero = FALSE,
  N_std = 5,
  N_dropNA = TRUE,
  chr_style = "Ensembl",
  rmv_chr = c("X", "Y", "MT"),
  on_ref_genome = TRUE,
  infer_eff_direction = TRUE,
  strand_ambig_filter = FALSE,
  allele_flip_check = TRUE,
  allele_flip_drop = TRUE,
  allele_flip_z = TRUE,
  allele_flip_frq = TRUE,
  bi_allelic_filter = TRUE,
  flip_frq_as_biallelic = FALSE,
  snp_ids_are_rs_ids = TRUE,
  remove_multi_rs_snp = FALSE,
  frq_is_maf = TRUE,
  indels = TRUE,
  drop_indels = FALSE,
  drop_na_cols = c("SNP", "CHR", "BP", "A1", "A2", "FRQ", "BETA", "Z", "OR", "LOG_ODDS",
    "SIGNED_SUMSTAT", "SE", "P", "N"),
  dbSNP = 155,
  check_dups = TRUE,
  sort_coordinates = TRUE,
  nThread = 1,
  save_path = tempfile(fileext = ".tsv.gz"),
  write_vcf = FALSE,
  tabix_index = FALSE,
  return_data = FALSE,
  return_format = "data.table",
  ldsc_format = FALSE,
  save_format = NULL,
  log_folder_ind = FALSE,
  log_mungesumstats_msgs = FALSE,
  log_folder = tempdir(),
  imputation_ind = FALSE,
  force_new = FALSE,
  mapping_file = sumstatsColHeaders,
  rmv_chrPrefix = NULL
)

Arguments

path	Filepath for the summary statistics file to be formatted. A dataframe or datatable of the summary statistics file can also be passed directly to MungeSumstats using the path parameter.
ref_genome	name of the reference genome used for the GWAS ("GRCh37" or "GRCh38"). Argument is case-insensitive. Default is NULL which infers the reference genome from the data.
convert_ref_genome	name of the reference genome to convert to ("GRCh37" or "GRCh38"). This will only occur if the current genome build does not match. Default is not to convert the genome build (NULL).
chain_source	source of the chain file to use in liftover, if converting genome build ("ucsc" or "ensembl"). Note that the UCSC chain files require a license for commercial use. The Ensembl chain is used by default ("ensembl").
local_chain	Path to local chain file to use instead of downlaoding. Default of NULL i.e. no local file to use. NOTE if passing a local chain file make sure to specify the path to convert from and to the correct build like GRCh37 to GRCh38. We can not sense check this for local files. The chain file can be submitted as a gz file (as downloaed from source) or unzipped.
convert_small_p	Binary, should non-negative p-values <= 5e-324 be converted to 0? Small p-values pass the R limit and can cause errors with LDSC/MAGMA and should be converted. Default is TRUE.
convert_large_p	Binary, should p-values >1 be converted to 1? P-values >1 should not be possible and can cause errors with LDSC/MAGMA and should be converted. Default is TRUE.
convert_neg_p	Binary, should p-values <0 be converted to 0? Negative p-values should not be possible and can cause errors with LDSC/MAGMA and should be converted. Default is TRUE.
compute_z	Whether to compute Z-score column. Default is FALSE. This can be computed from Beta and SE with (Beta/SE) or P (Z:=sign(BETA)*sqrt(stats::qchisq(P,1,lower=FALSE))). Note that imputing the Z-score from P for every SNP will not be perfectly correct and may result in a loss of power. This should only be done as a last resort. Use 'BETA' to impute by BETA/SE and 'P' to impute by SNP p-value.
force_new_z	When a "Z" column already exists, it will be used by default. To override and compute a new Z-score column from P set force_new_z=TRUE.
compute_n	Whether to impute N. Default of 0 won't impute, any other integer will be imputed as the N (sample size) for every SNP in the dataset. Note that imputing the sample size for every SNP is not correct and should only be done as a last resort. N can also be inputted with "ldsc", "sum", "giant" or "metal" by passing one of these for this field or a vector of multiple. Sum and an integer value creates an N column in the output whereas giant, metal or ldsc create an Neff or effective sample size. If multiples are passed, the formula used to derive it will be indicated.
convert_n_int	Binary, if N (the number of samples) is not an integer, should this be rounded? Default is TRUE.
impute_beta	Binary, whether BETA should be imputed using other effect data if it isn't present in the sumstats. Note that this imputation is an approximation (for Z & SE approach) so could have an effect on downstream analysis. Use with caution. The different methods MungeSumstats will try and impute beta (in this order or priority) are: log(OR) 2. Z x SE Default value is FALSE.
es_is_beta	Binary, whether to map ES to BETA. We take BETA to be any BETA-like value (including Effect Size). If this is not the case for your sumstats, change this to FALSE. Default is TRUE.
impute_se	Binary, whether the standard error should be imputed using other effect data if it isn't present in the sumstats. Note that this imputation is an approximation so could have an effect on downstream analysis. Use with caution. The different methods MungeSumstats will try and impute se (in this order or priority) are: BETA / Z 2. abs(BETA/ qnorm(P/2)) Default is FALSE.
analysis_trait	If multiple traits were studied, name of the trait for analysis from the GWAS. Default is NULL.
ignore_multi_trait	If you have multiple traits (p-values) in the study but you want to ignorwe these and instead use a standard named p-value, set to TRUE. By default is FALSE which will check for multi-traits.
INFO_filter	numeric The minimum value permissible of the imputation information score (if present in sumstats file). Default 0.9.
FRQ_filter	numeric The minimum value permissible of the frequency(FRQ) of the SNP (i.e. Allele Frequency (AF)) (if present in sumstats file). By default no filtering is done, i.e. value of 0.
pos_se	Binary Should the standard Error (SE) column be checked to ensure it is greater than 0? Those that are, are removed (if present in sumstats file). Default TRUE.
effect_columns_nonzero	Binary should the effect columns in the data BETA,OR (odds ratio),LOG_ODDS,SIGNED_SUMSTAT be checked to ensure no SNP=0. Those that do are removed(if present in sumstats file). Default FALSE.
N_std	numeric The number of standard deviations above the mean a SNP's N is needed to be removed. Default is 5.
N_dropNA	Drop rows where N is missing.Default is TRUE.
chr_style	Chromosome naming style to use in the formatted summary statistics file ("NCBI", "UCSC", "dbSNP", or "Ensembl"). The NCBI and Ensembl styles both code chromosomes as ⁠1-22, X, Y, MT⁠; the UCSC style is ⁠chr1-chr22, chrX, chrY, chrM⁠; and the dbSNP style is ⁠ch1-ch22, chX, chY, chMT⁠. Default is Ensembl.
rmv_chr	Chromosomes to exclude from the formatted summary statistics file. Use NULL if no filtering is necessary. Default is c("X", "Y", "MT") which removes all non-autosomal SNPs.
on_ref_genome	Binary Should a check take place that all SNPs are on the reference genome by SNP ID. Default is TRUE.
infer_eff_direction	Binary Should a check take place to ensure the alleles match the effect direction? Default is TRUE.
strand_ambig_filter	Binary Should SNPs with strand-ambiguous alleles be removed. Default is FALSE.
allele_flip_check	Binary Should the allele columns be checked against reference genome to infer if flipping is necessary. Default is TRUE.
allele_flip_drop	Binary Should the SNPs for which neither their A1 or A2 base pair values match a reference genome be dropped. Default is TRUE.
allele_flip_z	Binary should the Z-score be flipped along with effect and FRQ columns like Beta? It is assumed to be calculated off the effect size not the P-value and so will be flipped i.e. default TRUE.
allele_flip_frq	Binary should the frequency (FRQ) column be flipped along with effect and z-score columns like Beta? Default TRUE.
bi_allelic_filter	Binary Should non-biallelic SNPs be removed. Default is TRUE.
flip_frq_as_biallelic	Binary Should non-bi-allelic SNPs frequency values be flipped as 1-p despite there being other alternative alleles? Default is FALSE but if set to TRUE, this allows non-bi-allelic SNPs to be kept despite needing flipping.
snp_ids_are_rs_ids	Binary Should the supplied SNP ID's be assumed to be RSIDs. If not, imputation using the SNP ID for other columns like base-pair position or chromosome will not be possible. If set to FALSE, the SNP RS ID will be imputed from the reference genome if possible. Default is TRUE.
remove_multi_rs_snp	Binary Sometimes summary statistics can have multiple RSIDs on one row (i.e. related to one SNP), for example "rs5772025_rs397784053". This can cause an error so by default, the first RS ID will be kept and the rest removed e.g."rs5772025". If you want to just remove these SNPs entirely, set it to TRUE. Default is FALSE.
frq_is_maf	Conventionally the FRQ column is intended to show the minor/effect allele frequency (MAF) but sometimes the major allele frequency can be inferred as the FRQ column. This logical variable indicates that the FRQ column should be renamed to MAJOR_ALLELE_FRQ if the frequency values appear to relate to the major allele i.e. >0.5. By default this mapping won't occur i.e. is TRUE.
indels	Binary does your Sumstats file contain Indels? These don't exist in our reference file so they will be excluded from checks if this value is TRUE. Default is TRUE.
drop_indels	Binary, should any indels found in the sumstats be dropped? These can not be checked against a reference dataset and will have the same RS ID and position as SNPs which can affect downstream analysis. Default is False.
drop_na_cols	A character vector of column names to be checked for missing values. Rows with missing values in any of these columns (if present in the dataset) will be dropped. If NULL, all columns will be checked for missing values. Default columns are SNP, chromosome, position, allele 1, allele2, effect columns (frequency, beta, Z-score, standard error, log odds, signed sumstats, odds ratio), p value and N columns.
dbSNP	version of dbSNP to be used for imputation (144 or 155).
check_dups	whether to check for duplicates - if formatting QTL datasets this should be set to FALSE otherwise keep as TRUE. Default is TRUE.
sort_coordinates	Whether to sort by coordinates of resulting sumstats
nThread	Number of threads to use for parallel processes.
save_path	File path to save formatted data. Defaults to tempfile(fileext=".tsv.gz").
write_vcf	Whether to write as VCF (TRUE) or tabular file (FALSE).
tabix_index	Index the formatted summary statistics with tabix for fast querying.
return_data	Return data.table, GRanges or VRanges directly to user. Otherwise, return the path to the save data. Default is FALSE.
return_format	If return_data is TRUE. Object type to be returned ("data.table","vranges","granges").
ldsc_format	DEPRECATED, do not use. Use save_format="LDSC" instead.
save_format	Output format of sumstats. Options are NULL - standardised output format from MungeSumstats, LDSC - output format compatible with LDSC and openGWAS - output compatible with openGWAS VCFs. Default is NULL. NOTE - If LDSC format is used, the naming convention of A1 as the reference (genome build) allele and A2 as the effect allele will be reversed to match LDSC (A1 will now be the effect allele). See more info on this here. Note that any effect columns (e.g. Z) will be inrelation to A1 now instead of A2.
log_folder_ind	Binary Should log files be stored containing all filtered out SNPs (separate file per filter). The data is outputted in the same format specified for the resulting sumstats file. The only exception to this rule is if output is vcf, then log file saved as .tsv.gz. Default is FALSE.
log_mungesumstats_msgs	Binary Should a log be stored containing all messages and errors printed by MungeSumstats in a run. Default is FALSE
log_folder	Filepath to the directory for the log files and the log of MungeSumstats messages to be stored. Default is a temporary directory. Note the name of the log files (log messages and log outputs) are now the same as the name of the file specified in the save path parameter with the extension '_log_msg.txt' and '_log_output.txt' respectively.
imputation_ind	Binary Should a column be added for each imputation step to show what SNPs have imputed values for differing fields. This includes a field denoting SNP allele flipping (flipped). On the flipped value, this denoted whether the alelles where switched based on MungeSumstats initial choice of A1, A2 from the input column headers and thus may not align with what the creator intended.Note these columns will be in the formatted summary statistics returned. Default is FALSE.
force_new	If a formatted file of the same names as save_path exists, formatting will be skipped and this file will be imported instead (default). Set force_new=TRUE to override this.
mapping_file	MungeSumstats has a pre-defined column-name mapping file which should cover the most common column headers and their interpretations. However, if a column header that is in youf file is missing of the mapping we give is incorrect you can supply your own mapping file. Must be a 2 column dataframe with column names "Uncorrected" and "Corrected". See data(sumstatsColHeaders) for default mapping and necessary format.
rmv_chrPrefix	Is now deprecated, do. not use. Use chr_style instead - chr_style = 'Ensembl' will give the same result as rmv_chrPrefix=TRUE used to give.

Value

The address for the modified sumstats file or the actual data dependent on user choice. Also, if log files wanted by the user, the return in both above instances are a list.

Examples

# Pass path to Educational Attainment Okbay sumstat file to a temp directory

eduAttainOkbayPth <- system.file("extdata", "eduAttainOkbay.txt",
    package = "MungeSumstats"
)

## Call uses reference genome as default with more than 2GB of memory,
## which is more than what 32-bit Windows can handle so remove certain checks
## Using dbSNP = 144 for speed as it's smaller but you should use 155 unless
## you know what you are doing and need 144

is_32bit_windows <-
    .Platform$OS.type == "windows" && .Platform$r_arch == "i386"
if (!is_32bit_windows) {
    reformatted <- format_sumstats(
        path = eduAttainOkbayPth,
        ref_genome = "GRCh37",
        dbSNP = 144
    )
} else {
    reformatted <- format_sumstats(
        path = eduAttainOkbayPth,
        ref_genome = "GRCh37",
        on_ref_genome = FALSE,
        strand_ambig_filter = FALSE,
        bi_allelic_filter = FALSE,
        allele_flip_check = FALSE,
        dbSNP=144
    )
}
# returned location has the updated summary statistics file

neurogenomics/MungeSumstats documentation built on Aug. 10, 2024, 5:59 a.m.