User manual¶

Installation¶

Install the released package from PyPI:

pip install sushie
# or using uv
uv pip install sushie

To install from a source checkout:

git clone https://github.com/mancusolab/sushie.git
cd sushie
pip install .

or with uv:

uv sync

Data Preparation¶

Fine-mapping using individual-level data¶

To fine-map using individual-level data, SuShiE requires at least phenotype and genotype data specified with the option to specify covariates.

Although we highly recommend users to perform high-quality QC on their own genotype, phenotype, and covariate data, we implement following basic QCs in the software:

Remove subjects with N/A values from either phenotype or covariates data.
Remove SNPs that all subjects have N/A value.
Impute SNPs that partial subjects have N/A value based on two times allele frequencies.
Only keep subjects who have data in all the genotype, phenotype, and covariate data.
Only keep SNPs that are available in all the ancestries.
Adjust genotype data across ancestries based on the same reference alleles. Drop non-biallelic SNPs.
Remove SNPs that have minor allele frequency (MAF) less than 1% within each ancestry (users can change 1% with --maf).
Users also have an option to keep ambiguous SNPs (i.e., A/T, T/A, C/G, or G/C) by specifying --keep-ambiguous (Default is NOT to keep them).
Users have the option to perform rank inverse normalization transformation on the phenotype data.

See sushie.cli.process_raw for the implementation of these quality-control steps.

Fine-mapping using summary-level data (GWAS statistics)¶

To fine-map using summary-level data, SuShiE requires at least GWAS z statistics, sample sizes, and LD data. For LD data, users can provide individual-level genotype in PLINK1.9, PLINK2, VCF, or BGEN format and let SuShiE compute the LD matrix, or provide pre-computed LD matrix in tsv format.

Although we highly recommend users to perform high-quality QC on their own summary-level data, we implement following basic QCs in the software:

Remove SNPs with N/A values in GWAS.
Only keep SNPs that are available in all the ancestries.
Adjust GWAS and genotype data across ancestries based on the same reference alleles. Drop non-biallelic SNPs.
Remove SNPs (for LD computation) that have minor allele frequency (MAF) less than 1% within each ancestry (users can change 1% with --maf).
Users also have an option to keep ambiguous SNPs (i.e., A/T, T/A, C/G, or G/C) by specifying --keep-ambiguous (Default is NOT to keep them).

Testing Data¶

We provide example data in ./data/ folder to test out SuShiE. All the data are in three ancestries: 489 European individuals (EUR), 639 African individuals (AFR), and 481 East Asian individuals (EAS).

The genotype is the high-quality HapMap SNPs in some random gene 1M base-pair window, which contains 123, 129, and 113 SNPs for EUR, AFR, and EAS respectively in 1000G project. We provide genotype data in plink 1, vcf, and bgen 1.3 format.

Using ./data/make_example.py, we simulated phenotype data (2 causal QTLs, cis-SNP heritability: 0.5 and effect size correlation 0.8), random covariate data for each ancestry. The two QTL rsID are rs1886340 and rs10914958. It also outputs all.pheno file that row-binds simulated phenotype across ancestries, all.ancestry.index file that specifies ancestry index if using all.pheno, all.covar, and ./plink/all triplets, keep.subject file that specifies subjects to be included in the inference.

See Parameters for detailed format requirements.

Examples¶

SuShiE software is very easy to use, for it only has one command finemap. In this section, we walk through several examples of using SuShiE.

See Output files for detailed explanations of the generated files.

See Parameters for detailed explanations of command-line options.

We make a bash script ./misc/run_sushie.sh to show a more general working flow of running SuShiE.

If users still have questions, feel free to contact the developers.

Individual-level data example¶

Here are some examples for fine-mapping using individual-level data:

1. Two-Ancestry SuShiE¶

In this example, we perform two-ancestry SuShiE with covariates regressed out from both phenotypes and genotypes while updating the prior effect size covariance matrix during the optimizations.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --covar EUR.covar AFR.covar --output ./test_result

2. \(N\)-Ancestry SuShiE¶

In the example below, we perform single-ancestry SuShiE, which is equivalent to the SuSiE model (see the model reference).

cd ./data/
sushie finemap --pheno EUR.pheno --vcf vcf/EUR.vcf --covar EUR.covar --output ./test_result

Or three-ancestry setting:

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno EAS.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf vcf/EAS.vcf --covar EUR.covar AFR.covar EAS.covar --output ./test_result

3. Can I use other formats of genotypes?¶

Yes! SuShiE can take plink 1, plink 2, vcf, or bgen.

For VCF data, SuShiE uses a0=REF and a1=ALT and counts copies of the ALT allele. This differs from SuShiE 0.19 and earlier, which counted the REF allele for VCF input. Consequently, signed VCF weights from the newer convention have the opposite sign when their a0 and a1 columns are reversed relative to older results.

For plink 1, SuShiE read in the triplet (bed, bim, and fam) prefix.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --plink plink/EUR plink/AFR --output ./test_result

For plink 2, SuShiE reads the pgen, pvar, and psam prefix.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --plink2 plink2/EUR plink2/AFR --output ./test_result

Use --plink2-dosage when the plink 2 file should be read as dosage data.

For bgen data, users need to make sure that the latter allele shown up in the allele ids is the reference allele.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --bgen bgen/EUR.bgen bgen/AFR.bgen --output ./test_result

4. My data contains all the participants and I do not want to separate them¶

No problem! If all the subjects are in single phenotype, genotype, and covariate files. Users just need to use --ancestry-index command to specify a file that subject ID on the first column, and the ancestry index on the second column. The ancestry index has to start from 1 continuously to the total number of ancestry.

cd ./data/
sushie finemap --pheno all.pheno --plink plink/all --ancestry-index all.ancestry.index --output ./test_result

5. How about mega or meta SuShiE?¶

The software employs the function to run meta SuShiE and mega SuShiE by adding the parameter --meta or --mega.

We define the meta SuShiE as running single-ancestry SuShiE followed by meta analysis of the PIPs:

\[ \text{PIP}_{\text{meta}} = 1 - \prod_{i=1}^k(1 - \text{PIP}_{\text{ancestry i}}) \]

We define the mega SuShiE as running single-ancestry SuShiE on genotype and phenotype data that is row-wise stacked across ancestries.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --meta --output ./test_result

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --mega --output ./test_result

6. Let's estimate heritability, run CV, and make FUSION files!¶

SuShiE incorporates codes in limix to estimate the narrow-sense cis-heritability (\(h_g^2\)) by specifying --her.

SuShiE also has a function (--cv) to perform \(X\)-fold cross-validation (CV; --cv-num X) on the ancestry-specific prediction weights to compute the out-of-sample \(r^2\) between predicted and measured expressions with its corresponding \(p\)-value.

Specifically, we randomly (--seed [YOUR SEED]) and equally divide the dataset into X portions. We regard each portion as validation dataset and the rest four portions as training dataset. Then, we perform SuShiE on the training datasets for X times, and predict the expressions on the corresponding validation dataset. Last, we row-wise stack all X predicted expressions and compute the \(r^2\) with row-wise stacked and matched validation dataset.

With these two information (\(h_g^2\) and CV), we prepare R codes ./misc/make_FUSION.R to generate FUSION-format prediction weights, thus can be used in TWAS.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --cv --her --output ./test_result
Rscript ../misc/make_FUSION.R ./test_result ~

7. I don't want to scale my phenotype by its standard deviation¶

Fine-mapping inference sometimes can be sensitive to whether scaling the phenotypes and genotypes. SuShiE by default scales the phenotypes and genotypes by their respective standard deviations. However, if users want to disable it, simply add --no-scale to the command.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --no-scale --output ./test_result

8. I have my own initial values for the hyperparameters¶

SuShiE has three hyperparameters (see the model): the residual variance (\(\sigma^2_e\)) prior, the QTL effect size variance (\(\sigma^2_{i,b}\)) prior, and the ancestral effect size correlation (\(\rho\)) prior. SuShiE by default initializes them as 0.001, 0.001, and 0.8. If users have their own initial values, simply specify them with --resid-var, --effect-var, and --rho. Make sure the ancestry order has to match the phenotype file order.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --resid-var 2.2 2.2 --effect-var 1.2 3.4 --rho 0.2 --output ./test_result

By default, SuShiE will update \(\sigma^2_{i,b}\) and \(\rho\) during the optimization. If users want to disable it, add --no-update to the command line.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --resid-var 2.2 2.2 --effect-var 1.2 3.4 --rho 0.2 --no-update --output ./test_result

In addition, with --no-update, if users only specify --effect-var but not for --rho, --effect-var will be fixed during the optimizations while --rho will get updated, vice versa. In other words, if users want to fix both priors, they have to specify both at the same time or specify neither of them (in the latter case, fixing the default values 0.001 and 0.2 as the priors).

9. What if I assume no correlation across ancestries?¶

SuShiE features that it accounts for ancestral quantitative trait loci (QTL) effect size correlation (\(\rho\) in the model) in the inference, which is different from other SuSiE-extended multi-ancestry fine-mapping frameworks assuming no ancestral correlation (Joint SuShiE). However, it has the functions to make inference assuming no correlation across ancestries by simply specifying --no-update on the effect size covariance matrix and fixing the rho equal to zero --rho 0. With this, the effect size variance (\(\sigma^2_{i,b}\) in the model) will get updated while rho will not.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --no-update --rho 0 --output ./test_result

10. I want to improvise in post-hoc analysis¶

We understand the standard output files may not serve all users' post-hoc analysis. Therefore, we add the option to save all the inference results in *.npy file by specifying --numpy.

The *.npy files include SNP information, prior estimators, posterior estimators, credible set, PIPs, and sample size.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --numpy --output ./test_result

11. I seek to use GPU or TPU to make inference faster¶

SuShiE software uses JAX with Just In Time compilation to achieve high-speed computation. Jax can be run on GPU or TPU.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --platform gpu --output ./test_result

12. I want to use 32-bit precision¶

SuShiE uses 64-bit precision to assure an accurate inference. However, if users want to use 32-bit precision, they can specify it by having --precision 32.

Unless necessarily needed, we do not recommend to use 32-bit precision as it may cause non-positive semi-definite effect size covariance prior or decreasing ELBO, thus concluding the inference earlier.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --precision 32 --output ./test_result

13. I want to run fine-mapping on certain subjects¶

Users can use --keep command to specify a file that contains the subject IDs. As a result, the following fine-mapping inference only performs on the subjects listed in the file.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --keep keep.subject --output ./test_result

14. I want to assign the prior weights for each SNP¶

Users can use --pi command to specify a tsv file that contains the SNP ID and their prior weights. The weights will be normalized to sum to 1 before inference.

cd ./data/
sushie finemap --pheno EUR.pheno AFR.pheno --vcf vcf/EUR.vcf vcf/AFR.vcf --pi prior_weights --output ./test_result

Summary-level data example¶

Here are some examples for fine-mapping using summary-level data:

1. I want to perform fine-mapping on summary-level data and I provide individual-level reference panels for LD¶

cd ./data/
sushie finemap --summary --gwas EUR.gwas AFR.gwas --vcf vcf/EUR.vcf vcf/AFR.vcf --sample-size 489 639 --output ./test_result

2. I want to perform fine-mapping on summary-level data and I provide pre-computed LD matrix¶

cd ./data/
sushie finemap --summary --gwas EUR.gwas AFR.gwas --ld EUR.ld AFR.ld --sample-size 489 639 --output ./test_result

3. I want to only focus on SNPs with GWAS P values less than 5e-8 across `all` ancestries¶

cd ./data/
sushie finemap --summary --gwas EUR.gwas AFR.gwas --vcf vcf/EUR.vcf vcf/AFR.vcf --sample-size 489 639 --gwas-sig 5e-8 --gwas-sig-type all --output ./test_result

4. I want to only focus on SNPs between 34886700 and 35128637 on chromosome 1¶

cd ./data/
sushie finemap --summary --gwas EUR.gwas AFR.gwas --vcf vcf/EUR.vcf vcf/AFR.vcf --sample-size 489 639 --chrom 1 --start 34886700 --end 35128637 --output ./test_result

5. My GWAS data has different column names¶

cd ./data/
sushie finemap --summary --gwas EUR.gwas AFR.gwas --vcf vcf/EUR.vcf vcf/AFR.vcf --sample-size 489 639 --gwas-header CHR SNP BP A1 A2 STAT --output ./test_result

6. I want to add small number to diagonal of my LD matrix to make it positive definite¶

cd ./data/
sushie finemap --summary --gwas EUR.gwas AFR.gwas --ld EUR.ld AFR.ld --sample-size 489 639 --ld-adjust 1e-3 --output ./test_result

Parameters¶

Parameter	Type	Default	Example	Notes
`--summary`	Boolean	False	`--summary # will store as True`	Indicator whether to run fine-mapping on summary statistics. Default is False. If True, the software will need GWAS files as input data by specifying --gwas and need LD matrix by specifying either --ld or one of the genotype inputs. If False, the software will need phenotype data by specifying --pheno and genotype data by specifying --plink, --plink2, --plink2-dosage, --vcf, or --bgen.
`--pheno`	String	Required, no default	`--pheno EUR.pheno AFR.pheno`	Phenotype data. It has to be a tsv file that contains at least two columns where the first column is subject ID and the second column is the continuous phenotypic value. It can be a compressed file (e.g., tsv.gz). It is okay to have additional columns, but only the first two columns will be used. No headers. Use `space` to separate ancestries if more than two. SuShiE currently only fine-maps on continuous data.
`--plink`	String	None	`--plink plink/EUR plink/AFR`	Genotype data in plink 1 format. The plink triplet (bed, bim, and fam) should be in the same folder with the same prefix. Use `space` to separate ancestries if more than two. Keep the same ancestry order as phenotype's.
`--plink2`	String	None	`--plink2 plink2/EUR plink2/AFR`	Genotype data in plink 2 format. The pgen, pvar, and psam files should be in the same folder with the same prefix. Use `space` to separate ancestries if more than two. Keep the same ancestry order as phenotype's.
`--plink2-dosage`	String	None	`--plink2-dosage plink2/EUR plink2/AFR`	Dosage data in plink 2 format. The pgen, pvar, and psam files should be in the same folder with the same prefix. Use `space` to separate ancestries if more than two. Keep the same ancestry order as phenotype's.
`--vcf`	String	None	`--vcf vcf/EUR.vcf vcf/AFR.vcf`	Genotype data in vcf format. Use `space` to separate ancestries if more than two. Keep the same ancestry order as phenotype's.
`--bgen`	String	None	`--bgen bgen/EUR.bgen bgen/AFR.bgen`	Genotype data in bgen 1.3 format. Use `space` to separate ancestries if more than two. Keep the same ancestry order as phenotype's.
`--ancestry-index`	String	None	`--ancestry-index all.ancestry.index`	Single file that contains subject ID and their ancestry index. Default is None. It has to be a tsv file that contains at least two columns where the first column is the subject ID and the second column is the ancestry index starting from 1 (e.g., 1, 2, 3 etc.). It can be a compressed file (e.g., tsv.gz). Only the first two columns will be used. No headers. If this file is specified, it assumes that all the phenotypes across ancestries are in one single file, and same thing for genotypes and covariates data. It will produce errors if multiple phenotype, genotype, and covariates are specified.
`--keep`	String	None	`--keep keep.subject`	Single file that contains subject ID across all ancestries that are used for fine-mapping. It has to be a tsv file that contains at least one columns where the first column is the subject ID. It can be a compressed file (e.g., tsv.gz). No headers. If this file is specified, all phenotype, genotype, and covariates data will be filtered down to the subjects listed in it.
`--covar`	String	None	`--covar EUR.covar AFR.covar`	Covariates that will be accounted in the fine-mapping. It has to be a tsv file that contains at least two columns where the first column is the subject ID. It can be a compressed file (e.g., tsv.gz). No headers. All the columns will be counted. Use `space` to separate ancestries if more than two. Keep the same ancestry order as phenotype's. Pre-converting the categorical covariates into dummy variables is required. If the categorical covariate has `n` levels, make sure the dummy variables have `n-1` columns.
`--ld`	String	None	`--ld EUR.ld AFR.ld`	LD files that will be used in the fine-mapping. Default is None. Keep the same ancestry order as GWAS files. It has to be a TSV or compressed file (e.g., `.tsv.gz`). The header has to be the SNP name matching the GWAS data in `--gwas`. It can have fewer or more SNPs than the GWAS data, and the software will find the overlapping SNPs. Users must ensure that the LD and GWAS z statistics are computed using the same counting alleles.
`--chrom`	Integer	None	`--chrom 6`	Chromosome number to subset GWAS SNPs in the fine-mapping. Default is None. Value has to be an integer number between 1 and 22. The SNP position information from the first ancestry will be used for filtering. If this flag is specified, `--start` and `--end` must also be provided.
`--start`	Integer	None	`--start 1`	Base-pair start position to subset GWAS SNPs in the fine-mapping. Default is None. Value has to be a positive integer number. The SNP position information from the first ancestry will be used for filtering. If this flag is specified, --chrom and --end must also be provided.
`--end`	Integer	None	`--end 1000000`	Base-pair end position to subset GWAS SNPs in the fine-mapping. Default is None. Value has to be a positive integer number. The SNP position information from the first ancestry will be used for filtering. If this flag is specified, --chrom and --start must also be provided.
`--sample-size`	Integer	None	`--sample-size 489 639`	GWAS sample size of each ancestry. Default is None. Values have to be positive integer. Use 'space' to separate ancestries if more than two. The order has to be the same as the GWAS data in --gwas.
`--gwas-header`	String	chrom snp pos a1 a0 z	`--gwas-header CHR SNP BP A1 A0 Z`	GWAS file header names. Default is `chrom snp pos a1 a0 z`. Users can specify the header names for the GWAS data in this order.
`--gwas-sig`	Float	None	`--gwas-sig 5e-8`	The significance threshold for SNPs to be included in the fine-mapping. Default is 1.0. Only SNPs with P value less than this threshold will be included. It has to be a float number between 0 and 1.
`--gwas-sig-type`	String	at-least	`--gwas-sig-type all`	The cases how to include significant SNPs in the fine-mapping across ancestries. If it is 'at-least', the software will include SNPs that are significant in at least one ancestry. If it is 'all', the software will include SNPs that are significant in all ancestries. Default is 'at-least'. The significant threshold is specified by --gwas-sig.
`--L`	Integer	10	`--L 5`	Integer number of shared effects pre-specified. Larger number may cause slow inference.
`--pi`	String	"uniform"	`--pi ./prior_weights`	Prior probability for each SNP to be causal (\(\pi\) in model). Default is uniform (i.e., `1/p` where `p` is the number of SNPs in the region. It is the fixed across all ancestries. Alternatively, users can specify the file path that contains the prior weights for each SNP. The weights have to be positive value. The weights will be normalized to sum to 1 before inference. The file has to be a tsv file that contains two columns where the first column is the SNP ID and the second column is the prior weights. Additional columns will be ignored. For SNPs do not have prior weights in the file, it will be assigned the average value of the rest. It can be a compressed file (e.g., tsv.gz). No headers.
`--resid-var`	Float	1e-3	`--resid-var 5.18 0.2`	Specify the prior for the residual variance (\(\sigma^2_e\) in model) for ancestries. Values have to be positive. Use `space` to separate ancestries if more than two.
`--effect-var`	Float	1e-3	`--effect-var 5.21 0.99`	Specify the prior for the causal effect size variance (\(\sigma^2_{i,b}\) in model) for ancestries. Values have to be positive. Use `space` to separate ancestries if more than two. If `--no-update` is specified and `--rho` is not, specifying this parameter will only fix `effect-var` as prior through optimizations and update `rho`. If `--effect-var`, `--rho`, and `--no-update` all three are specified, both `--effect-var` and `--rho` will be fixed as prior through optimizations. If `--no-update` is specified, but neither `--effect-var` nor `--rho`, both `--effect-var` and `--rho` will be fixed as default prior value through optimizations.
`--rho`	Float	0.1	`--rho 0.05`	Specify the prior for the effect correlation (\(\rho\) in model) for ancestries. Default is 0.1 for each pair of ancestries. Use space to separate ancestries if more than two. Each rho has to be a float number between -1 and 1. If there are `N > 2` ancestries, `X = choose(N, 2)` is required. The rho order has to be `rho(1,2)`, ..., `rho(1, N)`, `rho(2,3)`, ..., `rho(N-1. N)`. If `--no-update` is specified and `--effect-var` is not, specifying this parameter will only fix `rho` as prior through optimizations and update `effect-covar`. If `--effect-var`, `--rho`, and `--no-update` all three are specified, both `--effect-var` and `--rho` will be fixed as prior through optimizations. If `--no-update` is specified, but neither `--effect-var` nor `--rho`, both `--effect-var` and `--rho` will be fixed as default prior value through optimizations.
`--no-scale`	Boolean	False	`--no-scale # will store as True`	Indicator to scale the genotype and phenotype data by standard deviation. Default is to scale. Specify `--no-scale` will store `True` value, and may cause different inference.
`--no-regress`	Boolean	False	`--no-regress # will store as True`	Indicator to regress the covariates on each SNP. Default is to regress. Specify `--no-regress` will store `True` value. It may slightly slow the inference, but can be more accurate.
`--no-update`	Boolean	False	`--no-update # will store as True`	Indicator to update effect covariance prior before running single effect regression. Default is to update. Specify `--no-update` will store `True` value. The updating algorithm is similar to EM algorithm or Empirical Bayes method that computes the prior covariance conditioned on other parameters. See the manuscript for more information."
`--max-iter`	Integer	500	`--max-iter 300`	Maximum iterations for the optimization. Larger number may slow the inference while smaller may cause different inference.
`--min-tol`	Float	1e-3	`--min-tol 1e-4`	Minimum tolerance for the convergence. Smaller number may slow the inference while larger may cause different inference.
`--threshold`	Float	0.95	`--threshold 0.9`	Specify the PIP threshold for SNPs to be included in the credible sets. It has to be a float number between 0 and 1.
`--purity`	Float	0.5	`--purity 0.5`	Specify the purity threshold for credible sets to be output. It has to be a float number between 0 and 1.
`--purity-method`	String	"weighted"	`--purity-method max`	Specify how to compute purity across ancestries. `weighted` is the sum of each ancestry's purity weighted by sample size, `max` is the maximum purity, and `min` is the minimum purity. Default is `weighted`.
`--ld-adjust`	Float	0	`--ld-adjust 1e-3`	The adjusting number to LD diagonal to ensure the positive definiteness. It has to be positive integer number between 0 and 0.1. Default is 0.
`--max-select`	Integer	250	`--max-select 100`	The maximum selected number of SNPs to calculate the purity. Default is 250. It has to be positive integer number. A larger number can unnecessarily spend much memory.
`--min-snps`	Integer	100	`--min-snps 50`	The minimum number of SNPs to fine-map. Default is 100. It has to be positive integer number. A smaller number may produce weird results.
`--maf`	float	0.01	`--maf 0.05`	Threshold for minor allele frequency (MAF) to filter out SNPs for each ancestry. It has to be a float between 0 (exclusive) and 0.5 (inclusive).
`--rint`	Boolean	False	`--rint`	Indicator to perform rank inverse normalization transformation (rint) for each phenotype data. Default is False (do not transform). Specify --rint will store 'True' value. We suggest users to do this QC during data preparation.
`--no-reorder`	Boolean	False	`--no-reorder`	Indicator to re-order single effects based on Frobenius norm of effect size covariance prior. Default is False (to re-order). Specify --no-reorder will store 'True' value.
`--keep-ambiguous`	Boolean	False	`--keep-ambiguous`	Indicator to keep ambiguous SNPs (i.e., A/T, T/A, C/G, or G/C) from the genotypes. Recommend to remove these SNPs if each ancestry data is from different studies or plan to use the inference results for downstream analysis with other datasets. Default is False (do not keep). Specify --keep-ambiguous will store 'True' value.
`--meta`	Boolean	False	`--meta # will store as True`	Indicator to perform single-ancestry SuShiE followed by meta analysis of the results. Specify `--meta` will store `True` value and increase running time. Specifying one ancestry in phenotype and genotype parameter will ignore `--meta`.
`--mega`	Boolean	False	`--mega # will store as True`	Indicator to perform mega SuShiE that run single-ancestry SuShiE on genotype and phenotype data that is row-wise stacked across ancestries. Specify `--mega` will store `True` value and increase running time. Specifying one ancestry in phenotype and genotype parameter will ignore `--mega`.
`--her`	Boolean	False	`--her # will store as True`	Indicator to perform heritability (\(h_g^2\)) analysis using limix. Specify `--her` will store `True` value and increase running time. It estimates \(h_g^2\) using codes in limix.
`--cv`	Boolean	False	`--cv 0.5 # will store as True`	Indicator to perform cross validation (CV) and output CV results (adjusted r-squared and its p-value) for future FUSION pipeline. Specify `--cv` will store `True` value and increase running time.
`--cv-num`	Integer	5	`--cv-num 6`	The number of fold cross validation. It has to be a positive integer number. Larger number may cause longer running time.
`--seed`	Integer	12345	`--seed 4321`	The seed for randomization. It can be used to cut data sets in cross validation. It can also be used to randomly select SNPs in the credible sets to calculate the purity. Default is 12345. It has to be positive integer number.
`--alphas`	Boolean	False	`--alphas # will store as True`	Indicator to output all the credible set results before pruning for purity including PIPs, \(\alpha\) (in model), whether in cs, across all \(L\). Default is False. Specify --alphas will store 'True' value and increase running time.
`--numpy`	Boolean	False	`--numpy # will store as True`	Indicator to output all the results in .npy file. Specify `--numpy` will store `True` and increase running time. .npy file contains all the inference results including SNP information, credible sets, pips, priors, posteriors, and sample size for users' own post-hoc analysis.
`--trait`	String	"Trait"	`--trait GENE_ABC`	Trait, tissue, gene name of the phenotype for better indexing in post-hoc analysis.
`--quiet`	Boolean	False	`--quiet # will store as True`	Indicator to not print message to console. Specify `--quiet` will store `True` value.
`--verbose`	Boolean	False	`--verbose # will store as True`	Indicator to include debug information in the log. Specify `--verbose` will store `True` value.
`--compress`	Boolean	False	`--compress # will store as True`	Indicator to compress all output tsv files in 'tsv.gz'. Specify `--compress` will store `True` value to save disk space. This command will not compress npy files.
`--platform`	String choices in `["cpu", "gpu", "tpu"]`	cpu	`--platform gpu`	Indicator for the JAX platform.
`--jax-precision`	Integer choices in `[32, 64]`	64	`--jax-precision 32`	Indicator for the JAX precision: 64-bit or 32-bit. Choose 32-bit may cause 'elbo decreases' warning.
`--output`	String	sushie_finemap	`--output folder/trait_name`	Prefix for output files.