Estimating local SNP-heritability¶

This page describes the steps to estimate local SNP-heritability from GWAS summary association data. In the first step, HESS computes the eigenvalues of LD matrices, and the squared projections of GWAS effect size vector onto the eigenvectors of LD matrices. In the second step, HESS uses the output from step 1 to obtain estimates of local SNP-heritability and their standard errors.

Step 1 - compute eigenvalues and squared projections¶

Running the tool¶

In this step, HESS computes the eigenvalues, and the squared projections of GWAS effect size vector onto the eigenvectors of LD matrices. The following code provides an example of how to perform this tep.

for chrom in $(seq 22)
do
    python hess.py \
        --local-hsqg <summary statistics> \
        --chrom $chrom \
        --bfile <reference panel in PLINK format for the specific chromosome> \
        --partition <genome partition file for the specific chromosome> \
        --out step1
done

In the command above, --local-hsqg tells HESS to estimate local SNP-heritability, and is used to specify the GWAS summary statistics data; --chrom is used to specify the chromosome number; --bfile is used to specify the reference panel for the corresponding chromsome; --partition is used to specify the genome partition file; --out is used to specify the prefix of the output file. For input file format, please refer to Input Format.

( Note: The above for loop can be parallelized by chromosome in case a computer cluster is available. In addition, users can specify the minimum minor allele frequency (MAF) of the SNPs used for estimation through the "--min-maf" flag. The default MAF threshold is 0.05. )

Interpreting the output¶

After executing the command above, 4 files will be created for each chromosome (i.e. 88 files for all 22 chromosomes in total), taking up ~10MB of space for the entire genome. The following is an example output obtained for chromosome 22.

step1_chr22.info.gz - contains the locus information, including start and end positions, number of SNPs, rank of LD matrices, and sample size

16050408        17674294        371     274     91273
17674295        18296087        419     306     89182
18296088        19912357        947     502     90231
...             ...             ...     ...     ...

step1_chr22.eig.gz - contains the eigenvalues of LD matrix (up to the rank of the LD matrix) at each locus, one line per locus

39.31792281  31.23990243  23.81549256  23.47296559  20.45343550  ...
48.73186142  26.95692375  25.32769526  22.11750791  20.55766423  ...
82.58157342  67.42588424  59.52766188  43.10471854  32.15181631  ...
...          ...          ...          ...          ...          ...

step1_chr22.prjsq.gz - contains the squared projections of GWAS effect size vector onto the eigenvectors of LD matrix (up to the rank of the LD matrix) at each locus, one line per locus

0.00008940  0.00001401  0.00013805  0.00009906  0.00007841  ...
0.00054948  0.00001756  0.00008532  0.00002303  0.00004706  ...
0.00008693  0.00005737  0.00070234  0.00008411  0.00004001  ...
...         ...         ...         ...         ...         ...

step1_chr22.log - contains helpful information for debugging, including number of SNPs, number of SNPs filtered, etc.

[INFO] Command started at: ...
[INFO] Command issued: ...
[INFO] Loaded ... partitins in ...
[INFO] Average window size is ...
[INFO] ... SNPs read from reference panel
                     ...
[INFO] Command finished at: ...

( Note: Log files are very useful in solving issues with running the software. Please include the log file in the email when you report any issue with HESS. )

Step 2 - estimate local SNP-heritability and standard error¶

Running the tool (without using total SNP-heritability)¶

In the second step, HESS uses output from step 1 across all chromosomes (step1_chr{1..22}.info.gz, step1_chr{1..22}.eig.gz, step1_chr{1..22}.prjsq.gz) to obtain local SNP-heritability estimates and their standard errors. The following command automatically looks for output from step 1 across all chromosomes with the prefix "step1" to obtain local SNP-heritability estimates.

python hess.py --prefix step1 --out step2

( Note: This is the recommended way of running HESS. However, if sample size of the GWAS is relatively small and total SNP-heritability is known, e.g. estimated from individual-level data, then one may use the script in the next section to obtain local SNP-heritability estimates. Also note that no for loop is needed here. )

Running the tool (using total SNP-heritability)¶

If sample size of the GWAS is relatively small, the standard error of total SNP-heritability estimates of HESS can be large, which could then lead to large standard error in the local estimates. However, if an estimate of total SNP-heritability is available, then one may use the following script to decompose the total SNP-heritability into each locus.

python hess.py --prefix step1 --tot-hsqg <total SNP-heritability> <SE> --out step2

Other available flags¶

--max-num-eig: Specifies the number of eigenvectors to use in the truncated-SVD for inverting the LD matrix, default 50
--min-eigval: Specifies the minimum eigenvalue cut off in truncated-SVD, default 1.0
--reinflate-lambda-gc: Specifies the genomic control factor that has been applied to the summary statistics, default 1.0
--gwse-thres: A threshold to cap the standard error of total SNP-heritability estimate

( Note: We recommend to run HESS on GWAS summary statistics that has not been applied the genomic control factor, which could results in downward bias. )

We provide a simple script (misc/estimate_lambdagc.py) to infer the genomic control factor to reinflate the summary statistics if it is unknown. The following script should be executed after step 1 completes.

python misc/estimate_lambdagc.py --prefix step1

Interpreting the output¶

After step 2, 2 files will be created. These include

step2.txt: contains local SNP-heritability estimates for all loci across all chromosomes. The columns are chromosome number, locus start position, locus end position, number of SNPs in locus, number of eigenvectors used in truncated-SVD, estimates of local SNP-heritability, estimated variance of the local SNP-heritability. We also report standard error and test statistics.

chr start       end         num_snp k   local_h2g   var         se          z           p
1   10583       1892607     1286    50  0.0001293   1.2819e-08  0.00011322  1.142       0.12673
1   1892607     3582736     3045    50  0.0003232   2.7102e-08  0.00016463  1.9632      0.024809
1   3582736     4380811     1622    50  6.7587e-05  1.282e-08   0.00011323  0.59692     0.27528
1   4380811     5913893     3790    50  3.1445e-05  2.6111e-08  0.00016159  0.1946      0.42285
... ...         ...         ...     ... ...         ...         ...         ...         ...
22  46470495    47596318    2444    50  0.00018004  2.0125e-08  0.00014186  1.2691      0.1022
22  47596318    48903703    2997    50  8.389e-05   2.4784e-08  0.00015743  0.53287     0.29706
22  48903703    49824534    3773    50  6.9487e-06  2.5931e-08  0.00016103  0.043151    0.48279
22  49824534    51243298    2789    50  0.00015227  2.4024e-08  0.000155    0.98241     0.16295

step2.log: contains useful information for debugging

[INFO] Command started at: Tue, 03 Oct 2017 21:01:34
[INFO] Command issued:
    /u/project/pasaniuc/shihuwen/software/hess/hess.py \
        --prefix ... \
        --out ...
[INFO] Loaded results for ... loci from step 1
[INFO] Using 4381458 SNPs with average sample size ...
[INFO] Re-inflate the summary statistics with lambda_gc: ...
[INFO] Total SNP-heritability estimate: ...
[INFO] Command finished at: ...