Pre- and post-genotype filtration protocol to improve the variant imputation metrics with further quality control

doi:10.21203/rs.3.pex-2088/v1

Method Article

Pre- and post-genotype filtration protocol to improve the variant imputation metrics with further quality control

https://doi.org/10.21203/rs.3.pex-2088/v1

This work is licensed under a CC BY 4.0 License

This protocol has been posted on Protocol Exchange, an open repository of community-contributed protocols sponsored by Nature Portfolio. These protocols are posted directly on the Protocol Exchange by authors and are made freely available to the scientific community for use and comment.

Version 1

posted

You are reading this latest protocol version

The search for loci linked to diseases or drug targets is being studied by genotyping variants. Missing variants can be imputed, but the standard filtration methods applied during quality control(QC) reduced their number for downstream analyses.There is no standard protocol to retain the maximum of variants with reliable quality. To counteract this loss, we first limited the stringency of MAF SNP pre-filtration (MAF-DS), which improved post-imputation metrics with the 1000 Genomes (inc. INFO based on allele frequencies). Second, our two-step post-filtration strategy (1)computes a confident MAF threshold (MAF-DH), (2)filters with the quality score >0.3 and gains ~18% of reliable markers compared to conservative methods (single-step INFO>0.8) and confidence vs. non-conservative methods (single-step INFO>0.3).During the pre-and post-filtration steps, the MAF and allele accuracies were validated against a public database for further QC.The protocol is also applicable to any study since the metric MAFs are adapted to the sample set.

Computational biology and bioinformatics

1000 Genomes (1kGP)

accuracy

AGR

AGVP

allele frequency

Alzheimer’s disease

biomarkers

CAAPA

computational biology and bioinformatics

confidence

correlation

database

drug

filtration

fine mapping

genetics

genome

genotyping

GLIMPSE

gnomAD

GWAS

haplotype

HapMap

HRC

IMPUTE

imputation

information

low-coverage genotype-by-sequencing (lc-GBS)

low-pass genome sequencing (LP-GS)

MAF

marker

meta-analyses

metrics

microarray

PCR

phasing

post-filtration

pre-filtration

quality control (QC)

rare variants

reference genome

repositioning

sequencing

SHAPEIT

SNP

target

TOPMed

UK10K.

Genetic imputation of markers infers additional variants from a reference genome at unobserved genotypes to investigate loci responsible for diseases. The search for biomarkers in Mendelian inherited diseases by genetic mapping of highly polymorphic sites paved the way for genome-wide linkage screens of complex traits^1-3. The physical map of the human genome built with sequence tag sites (STSs)^4,5 allowed nearly 200 microsatellite tandem repeats to be reliably PCR-amplified to pursue linkage analyses in different pathologies^6,7. The number of validated microsatellite markers with the successive releases of the human book of life was subsequently improved to more than 400 for routine genome-wide scans^8,9,10,11. The loci identified in different pathologies were further refined by using additional polymorphic markers through linkage disequilibrium in family-based association studies^12-14.

Later, the identification of single-nucleotide polymorphisms (SNPs) from diverse sources^15-20 densified the map of the genetic variants. Large screening of variants allowed high-throughput genotyping with SNP arrays^21,22 in genome-wide linkage scans to unravel more loci linked to diseases ^23-25. Concurrently, whole-genome association studies (GWAS) revealed genetic markers in human disorders^26-29, and their investigation was enriched by imputation³⁰.

Following microarray-based genotyping, the imputation of untyped variants from a reference genome generated more abundant SNPs³¹. In return, it improved the power of association analyses^32-34. Anderson et al. (2008) demonstrated a gain in power to detect an association after imputing SNPs (0.05% < MAF < 0.1) of allelic odds ratios 1.6-2 (effects size)³³. The imputation of SNPs also increased the number of markers to fine map genetic loci identified by GWAS to narrow down the regions of interest more precisely^25,35-37.

Imputed genotyped individuals thus revealed variants of disease-susceptibility genes not present on the array. For instance, Jonsson et al³⁸ discovered by imputation in the APP gene the rare variant (rs63750847-A) that protects against Alzheimer’s disorder (AD). They also confirmed that a rare mutation in the TREM2 gene (rs75932628-T) was associated with AD by replicating their findings in cohorts imputed with the 1000GP panel (1kGP) from phase 1 (1000GP1)³⁹.

GWAS from different cohorts studied in meta-analyses showed similar statistical power to the pooling of individuals from different genome-wide association datasets ^14,40. Imputation improved the power of meta-analyses by increasing the number of shared variants across the genome in the joint summaries of genome-wide association studies (GWASs)⁴¹.

Druggable and candidate loci present in one pathology found in another clinical disorder can also be investigated from GWAS for the repurposing of drugs^42,43. Gene-based statistics derived from an entire set of GWAS data⁴⁴ with pathway studies⁴⁵ and meta-analyses are further used for finding potential drug targets⁴⁶. In addition, disease risk genes such as the APOE 4 locus identified in AD using genetic linkage mapping⁴⁷ confirmed by genetic associations^48-50 and meta-analyses⁵¹ can also be screened for drug repositioning⁵². Meta-analyses and imputation of variants can therefore identify or re-ascertain SNPs in genes associated with traits such as AD (TREM2, APOE)⁵³ that will inform for the therapy or the repurposing of drugs⁵⁴. Highly polymorphic genes such as HLA (hman leucocyte antigens) imputed to a multi-population panel with the haplotype graph HLA*IMP:02⁵⁵ could recover allelic variants genetically associated with common diseases³. Today, imputation is also studied in the determination of polygenic risk scores (PRSs) in human disorders⁵⁶.

Early GWAS imputation used the reference genome panel from phase 2 of the international HapMap^31,57,58 to predict the variants missing on a genotyping array. Phase 1 of the 1000 Genomes Project (1000GP1) subsequently released by low-coverage and exome sequencing data from the 1,092 individuals²⁰ enabled the characterisation of low-frequency and rare variants. Phase 3 of the 1000GP (1000GP3) expanded the number of individuals to 2,504, thus increasing the number of total and rare variants⁵⁹. New structural variants were further uncovered with the 1000GP3⁶⁰ and the genome of the Netherlands project (GoNL)⁶¹. The 1000GP3 was then integrated into the large Haplotype Reference Consortium (HRC) panel ⁶², while the re-phased 1000G reference Panel 1 (v3)⁶³ combined with the UK10K panel⁶⁴ enhanced the findings and accuracy of imputed rare variants⁶⁵. The larger and multi-ethnic imputation panel from NHLBI (Trans-Omics for precision medicine - TOPMed) was recently deeply sequenced (WGS 38x) in approximately 97,256 individuals (TOPMed freeze 8 reference panel)^66,67. The 2,504 1kGP was further sequenced to a higher depth coverage of 30x with an additional 698 subjects⁶⁸. Other genome aggregation databases from numerous studies (gnomaD v2 inc. ExAC and v3) also provide low-frequency variants^69-71.

Vergara et al. (2018)⁷² reported that the average density of variants imputed across chromosomes was higher in the 1000GP3 panel compared to the HRC panel. Currently, the 1000GP3 combined with ~2,100 African samples also constitutes the African Genome Resource (AGR) reference panel at the Sanger centre (Sanger imputation server : https://imputation.sanger.ac.uk/?about=1#referencepanels). In addition to the existing African genome variation project (AGVP)⁷³, Mathias et al. 2016 mentioned the necessity for combined African datasets. They suggested the expansion of data for the H3 Africa Consortium⁷⁴ with the genomic variations produced by the Consortium on Asthma among African-Ancestry Populations in the Americas (CAAPA), also available on the Michighan server⁷⁵. The reference panels of CAAPA (dbGAP with accession ID:phs001123.v1.p1)^75,76 along with the 1000GP3 and the HRC reference panels could be joined to reveal additional rare variants. Taliun et al. 2021⁶⁷ found by imputation some of the very rare variants in Africans with the TopMed project, which includes the Chronic Obstructive Pulmonary Disease Cohort (COPD)⁷⁷.

Statistical models for imputation have been applied using Beagle^78-81, IMPUTE⁸² and MACH⁸³ with comparable accuracy^84,85. Both IMPUTE and MaCH applied the hidden Markov model (HMM) while Beagle previously relied on a graphical model of the haplotype set is now based on HMM⁸⁴.

Re-phasing of the 1000 Genomes reference panel using SHAPEIT2 against the 1000GP3 improved haplotyping and the imputation of rare variants⁶³. SHAPEIT2 also allows pre-phasing of the GWAS samples for a better ordering of the markers and ameliorates the performance of imputation⁸⁶. Lately, newer versions of SHAPEIT⁸⁷ and IMPUTE⁸⁸, both based on linkage disequilibrium (LD), have been further developed with the positional Burrows–Wheeler transform (PBWT)⁸⁹. Other methods of imputation are being developed to recover rare variants from genotyping-by-sequencing methods (GBS) (BEAGLE v4.1⁸⁴, GENEIMP v1.3⁹⁰, GLIMPSE⁹¹, LOIMPUTE v0.1.1⁹², QUILT⁹³ or STITCH⁹⁴).

Currently, in the absence of consensus for the quality control of imputation, the methods applied are mainly based on the standards used for GWAS. There is no standard protocol to obtain a balance between retaining enough variants and maintaining sufficient good quality before (pre-filtration) and after imputation (post-filtration) without stringent conditions. Here, we have established a procedure to keep markers that are valuable for improving imputation metrics⁹⁵.

Our workflow consists of pre- and post-imputation filtrations approaches that are different from conservative methods. After genotyping a) at pre-filtration, we considered the allele frequencies of the sample set (MAF-DS, MAF Dataset) instead of a stringent or arbitrary MAF b) and a 2-step post-filtration strategy after imputation with (1) a specific MAF to compute (MAF-DH, MAF Dosage representing 1.5 Heterozygous) and (2) a determined minimum imputation quality score (INFO). A database of control samples was used to compare the MAFs and alleles for a given SNP of similar number and population origin.

In our experience, the post-filtration approach with the specific MAF (MAF-DH) in the first step maintained SNPs with confidence above the INFO quality score (information) of 0.3. Only 1% of SNPs remained below this score and were further eliminated in the second step. Previous publications reported the possible role of MAFs and quality scores in improving imputation accuracy and quality⁷⁶. The number of heterozygous individuals who would be represented in at least in 1,000 subjects was in accordance with the number evoked by Coleman et al. 2016⁹⁶. We also found that the properties of the region studied applied to the genome and that applying our procedure reduced information loss⁹⁵. We defined guidelines for pre-filtrations to impute SNPs and post-filtration of imputed SNPs before downstream analyses in search of loci.

Our approach can be customised to any study since the MAF metric depends on the size of the sample set (MAF-DS) and on the MAF-DH, representing 1.5 heterozygous individuals calculated post-imputation from the imputed dosage (likely genotype). The method would be applicable to other panels, such as TopMed, especially in the East and South Asian populations where the imputation quality did not reach 0.8 at MAF 1E-04^66,67. The protocol could be extended to impute SNVs absent from genotyping arrays of animals^97,98 and plant breedings⁹⁹ and possibly to different biological tissues, including cell-free foetal DNA in maternal blood¹⁰⁰.

The imputation of variants remains cost-efficient compared with whole-genome sequencing (WGS)^72,101,102 even in fine-mapping of large sample sizes²⁵. Recently, the combination of sequenced individuals with the imputation of additional samples from GWAS increased the power of association analyses to reveal a hundred new signals in anthropometric traits¹⁰³. The imputation from low coverage genotyping-by-sequencing (Lc-GBS)¹⁰⁴, also referred to as low-pass genome sequencing (LP-GS)^91,105-107, which is also cost-efficient, generated a large number of markers in different studies^100,108 and recently proved to be performant⁹¹.

Depending on the precision from the latest platform used (genotyping-by-sequencing e.g., GBS or LP-GS) the first step of post-filtration by MAF-DH would generate a certain quality of imputation. The latter could be improved by the second step of post-filtration with the threshold determined more specifically after reporting the imputation performance score (e.g., INFO in IMPUTE or aggregated r² in GLIMPSE) as a function of the MAFs.

Hardware

· Unix machine CentOS 6.5, Kernel 2.6.32-431.29.2.el6.x86_64, Cluster Slurm 2.6.9 with 8 nodes, Mysql 5.6.

Files

· The empirical sample dataset for 1,031 individuals with the full description of the samples (37% AMR, 34.3% EUR, 25.1% EAS, 2.95% AFR and 0.65% SAS) is available at

http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/working/20130606_sample_info/20130606_sample_info.xlsx in pedigree (ped files) and mapping (map files) formats from https://mathgen.stats.ox.ac.uk/genetics_software/shapeit/shapeit.html (SHAPTEITv2.r790.RHELS_5.4)

· The NCBI reference dbSNP of the human build genome 137 based on GRCh37 assembly 37.4 (NCBI dbSNP B37.p13) was inserted in Mysql database (db_DBSNP_137_NCBI).

The NCBI files are available from

BED files : http://ftp.ncbi.nih.gov/snp/organisms/human_9606/BED/
VCF files by Chromosomes: ftp://ftp.ncbi.nih.gov/snp/organisms/human_9606/VCF/ByChromosome/ https://ftp.ncbi.nih.gov/snp/organisms/human_9606/VCF/

· The reference panel for imputation is based on 1,092 individuals from phase 1 (1000 GP1) (35% EUR, 26% ASN, 22% AFR, 17% AMR) :

http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20110521/. The genomes available in IMPUTE https://mathgen.stats.ox.ac.uk/impute/data_download_1000G_phase1_integrated_SHAPEIT2_16-06-14.html include the phased sequences of the 1000GP against 2,141 SNP array haplotypes of the same samples and additional individuals with the coordinates of the NCBI build 37 (hg19)⁶³ :

http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/phase1/analysis_results/shapeit2_phased_haplotypes/

Software

· R package¹⁰⁹, Bioconductor 3.2 version R 3.2.4 in the integrated development environment

· RStudio 1.3.959 (RStudio PBC, Boston, MA, USA) https://www.r-project.org/ for statistical analyses

· Whole-genome data analysis PLINK (v 1.07) toolset^110,111 : https://zzz.bwh.harvard.edu/plink/download.shtml

https://zzz.bwh.harvard.edu/plink/dist/plink-doc-1.07.pdf

PLINK 1.9 (Plink2) can be downloaded freely from https://www.cog-genomics.org/plink2 with the latest version of plink 2.00 alpha https://www.cog-genomics.org/plink/2.0/

· SHAPEIT2 v2.r790.RHELS_5.4^86,112

can be accessed from https://mathgen.stats.ox.ac.uk/genetics_software/shapeit/shapeit.html

The latest versions of SHAPEIT3 https://jmarchini.org/shapeit3/¹¹³ and SHAPEIT4 are also available at https:/odelaneau.github.io/shapeit4^87,114 for larger sample sets.

· IMPUTE2^30,82,115 can be downloaded from https://mathgen.stats.ox.ac.uk/impute/impute_v2.html.

Later versions of the programs exist and should further improve the protocol, IMPUTE4 is for larger sample datasets (~500,000) : https://www.dropbox.com/sh/k6b34fzw9w4s8bg/AAA65aF5l2oj_AT9iDLgKCv9a?dl=0 and IMPUTE5 is based on PBWT⁸⁸. In sequencing diverse programs are available for phasing and imputation such as GLIMPSE⁹¹.

· SNPTestV2^30,85,116 is freely available at https://www.well.ox.ac.uk/~gav/snptest/#download (SNPTEST v.2.5 .2) https://mathgen.stats.ox.ac.uk/genetics_software/snptest/old/snptest.html

A - PRE-IMPUTATION

1| Collect the sample dataset

The genotypes from 1,031 individuals are obtained in ped files and map files formats for the 2Mb region of chromosome 20 from the file source SHAPEIT2 (v2.r790.RHELS_5.4) (Figure 2¹¹⁷).

Problem 1

The coordinates for the variants can change.

Possible reason

The coordinates depend on the version of the human genome.

Solving problem 1

To impute samples, use a genotyping chip based on the same version as the imputation reference panel.

Problem 2

Rare variants are mis-clustered.

Possible reason

Clusters of rare variants may not be well defined or may be missed in GenomeStudio by the GenCall algorithm, which is more adapted to creating clusters of more common variants¹¹⁸, especially in smaller cohorts of less than 1,000 or 10,000 individuals.

Solving problem 2

To create better clusters of the rarest variants in GenomeStudio after genotyping, use the .idat and the sample files of the study genotypes if they are available from the laboratory that performed the genotyping. Guo et al. (2014) described how to optimise the classification of rare variants in GenomeStudio. We further suggest importing the .idat and the sample files from a diversity panel (e.g. HGDP¹¹⁹ or TopMed) genotyped on an appropriate microarray chip. This would increase the number of rare variants to better capture and define their clusters by chip. It should also be possible to download a sample file from an existing study of individuals carrying rare variants to increase the number of subjects to create larger batches of these SNPs to cluster them from the same centre (see TROUBLESHOOTING).

In GBS or LP-GS to improve sensitivity of called genotypes procedures recommend to use joint variants calling, and also both strands¹⁰⁷ and at times orthogonal validation when justified^120,121.

2| Compute the different quality control criteria

Quality control (QC) is performed with PLINK (v 1.07)¹¹⁰before phasing and imputation.

Compute with PLINK or one’s own software the success rate of individuals, heterozygosity, the search for duplicates and relatedness, Hardy-Weinberg equilibrium (HWE), and the success rates of SNPs and MAFs. Perform the population stratification by PCA with EIGENSTRAT (EIGENSOFT v4.2 : http://www.hsph.harvard.edu/alkes-price/software/

https://alkesgroup.broadinstitute.org/EIGENSOFT/)¹²²_.

Standard QC procedures include retaining SNPs with a Hardy-Weinberg p>1E-06 and SNPs with a call rate ≥99%, but when phenotypic status is available, SNPs with call rate differences are removed. Monomorphic markers are excluded during the pre-filtration based on the MAF.

Sample call rates >= 95% are also retained, duplicate or related samples in GWAS and subjects with gender mismatches between their genotypes and the sample files (when data from chromosome X are available) are excluded, and samples with heterozygosity >+ 3SD from the sample mean heterozygous rate and PCA outliers are also removed for GWAS.

3| Compute the MAF of the sample set genotyped (MAF-DS) for variant pre-filtration

Calculate the MAF of the sample set, MAF-DS (MAF-dataset) to retain the variants >= MAF of the sample studied instead of filtering the SNPs with a stringent filtration such as MAF < 0.01.

The MAF from the sample set, MAF-DS, is 5E-04, which is identical to the MAF (gmaf) from the NCBI database used as a control (step 6).

4| Variant pre-filtration

Autosomal SNPs are usually removed with a low call rate <99% and HWE <1E-06. In this dataset, SNPs with a low HWE correspond to a small number of common and high allelic frequencies. Call rates below 99% include only 5 SNPs with rare frequencies. The main cause for removal of the SNPs during pre-filtration would thus be the MAF stringency selection criteria, as in most QC, e.g. MAF <0.01. Therefore, in our protocol, the variants are filtered with the MAF of the sample set, MAF-DS (i.e., not performed below the MAF of the sample set 5E-04), limiting the exclusion of markers while pre-filtering the monomorphic variants. This can be done for any study. The correlation between the MAFs of the SNPs and the control database after pre-filtration was verified in step 6.

5| Individual filtration

Following pre-filtration of the variants, QC is performed on the individuals. The lowest success rate remains between 97%-99% for 4 individuals both before and after elimination of the variants. In this sample set, the individuals are thus all retained following genotyping, as their pass quality scores are in the range used by most GWASs.

6| Variants

6a| SNP control database (Figure 2)

(i) The control database needs to be of similar origin and sample size to the samples studied to compare the MAF for each SNP, so we suggest either a public reference or a designed combination (combo) database. In this example, the number of individual genotypes (1,031 IDs) is close to the NCBI MAFs dbSNPB137 based on 1,089 subjects from the 1000GP1. Thus, the minimum MAF in dbSNPB137 was 5E-04, identical to that of the sample set used referring to at least one heterozygous genotyped individual. The MAF from the 1000GP in NCBI was determined from the global minor allele frequencies (gmaf)²⁰.

(ii) Upload the NCBI database dbSNP of the human genome build B137 based on GRCh37 assembly 37.4 (dbSNP B137), e.g. , in Mysql Enterprise 5.6.13, or query a public online database.

(iii) Extract the list of SNPs with their MAFs from the database with their alleles.

(iv) Compare the MAFs of each SNP genotyped computed in step 2 against the MAFs of the same SNPs in the NCBI control database (Figure 2.1); similarly, this can be done for the alleles as per Figure 2.2.

The same control database is used to compare the MAFs of the genotyped variants and to compare the MAFs of the imputed variants as a control of accuracy (step 9).

Problem

Discordance is present between the names and coordinates of the SNPs genotyped with the reference control database.

Possible reason

The coordinates of the variants can change depending on the version of the human genome.

Solving the problem

Use a genotyping chip or genotype data of the study based on the same version as the MAF genome control database.

6b| MAF clusterisation

As previously shown, the variants are clustered in different levels of non-overlapping MAFs, such as null (0-1E-04], very rare (1E-04-1E-03], rare (1E-03-1E-02], low (1E-02-5E-02], common (5E-02-1E-01] and high (1E-01-5E-01] frequencies. The very rare and rare classes were subdivided into 2 bins: (1E-04-5E-04] and (5E-04-1E-03] for the very rare MAFs, (1E-03–5E-03] and (5E-03-1E-02] for the rare MAFs⁹⁵.

(i) Optional step: The classification can serve for the comparison of the post-imputation results under stringent vs. non-stringent pre-filtration conditions (Table 1, Figures 2.3, 2.4, 2.5, Table 2, Step 11 not included in Figure 1).

(ii) The classification of the MAF allows the reporting of the imputation quality score aggregated by MAF (Figure 2.6).

B - IMPUTATION

7| Phase and imputation

7a| Phase

Phase the genotypes from the pedigree files with SHAPEIT2 (v2.r790.RHELS_5.4).

7b| Imputation

Perform the imputation of the phased haplotypes with IMPUTE2 against the reference genome 1000GP1.

Optional step: Comparison between the two different filtration conditions of the study genotypes. In this case, phasing and imputation should be run in both conditions with the same seed as previously described⁹⁵. In sequencing phasing and imputation can be carried out with the same program such as GLIMPSE⁹¹.

Problem

Some variants on the array might be absent from the reference panel used for imputation.

Possible reason

The coordinates and names for the variants on the genotyping array might be different from the version of the human reference genome. In addition, some manufactured arrays have their own sets of SNPs; only 13 SNP study genotypes could not be found on the imputation reference panel but could be inferred (labelled as type 3) and were correctly imputed with the option available in IMPUTE2 https://mathgen.stats.ox.ac.uk/impute/impute2_overview.html.

Solving the problem

To impute use an imputation reference panel that is based on the same version as the genotyping chip; otherwise, some genotyped SNPs could differ in terms of coordinates, names, alleles but also MAFs.

C - POST-IMPUTATION

8| Post-imputation

(A) Calculate the imputed MAFs

Calculate the MAFs for each variant imputed with SNPTEST v2.5.2. https://www.well.ox.ac.uk/~gav/snptest/#data_summaries.

(B) Compute the MAF threshold post-imputation (MAF-DH) and check the standard metrics.

The MAF-DH defines the MAF computed for the imputed genotype probabilities ('dosages') that represents > 1 heterozygous individual in the sample set.

The MAF that shows between 1 and 2 individuals (1.5) in this example, including 1,032 individuals, is 7.5E-04 (MAF-DH, Dosage1.5Heterozygous), which is above the MAF of the sample set, MAF-DS (5E-04).

(C) Other controls of imputation are available in IMPUTE2, including the overall concordance score between original and imputed genotypes⁸⁵, the metric r2_type0 score for the accuracy of each imputed genotype (square correlation between imputed 'dosages' (expected genotypes) and the true genotypes^80,83,85,105 i.e., Input genotypes in IMPUTE2 : https://mathgen.stats.ox.ac.uk/impute/impute_v2.html#concordance_tables) with the quality scores of imputation, i.e., INFO metric scores (information) in IMPUTE2 of each variant. The HWE can be computed. Variants that deviate from HWE can be post-filtered of flagged for GWAS analyses.

9| Control the MAF and alleles imputed

(i) Extract the MAF of each variant imputed with their alleles.

(ii) Use the control database in step 6 to compare the MAFs of the imputed variants as a control of accuracy to complete the QC criteria of imputation (Figure 2.3). The alleles can also be checked by comparison for consistency.

Problem

The coordinates or names of the imputed variants differ from the coordinates or names of the control variants from the MAF database.

Possible reason

The coordinates and names of the SNPs change depending on the version of the human genome.

Solving the problem

A MAF control database of the same human genome build version is used for the imputation reference panel. In our case, we compared each imputed variant to the NCBI database. A curation was necessary after imputation to compare the MAFs between the imputed variants and the controls from NCBI when the coordinates or SNPs names differed, as the released databases were slightly different.

Very few imputed SNPs were thus curated by hand in the dataset to allocate the correct rs number for either duplicates, unreferenced SNPs, i.e., missing rs number from dbSNP, merged SNPs or rs numbers that had changed records from a polymorphism to a deletion at the same position in dbSNPs. For an optimal control when possible, use a curated control database.

10| Post-filtration in 2 steps

(A) Filtration with the MAF-DH threshold post-imputation

The imputation generates MAFs (1E-04) below the MAF-DS (5E-04), because it is based on probabilistic genotype predictions, and these variants should be eliminated. The MAF-DH threshold computed in step 8 to post-filter the MAFs imputed at 7.5E-04 eliminates the allele dosage that also represents less than one heterozygous imputed genotype.

The MAF-DH (7.5E-04) thus captures more than one heterozygous individual, improving the confidence of the marker and simultaneously excluding the monomorphic variants. This post-filtration step produces an average INFO score of 0.64 (Figure 2.6) within the MAF range 7.5E-04-1E-03 and less than 1% of SNPs with INFO scores < 0.3, which need to be removed.

(B) Post-filtration with information score

(i) It is necessary to finally post-filter at an INFO score of 0.3 to eliminate the small number of markers (~1%) with poor quality. Filtering with a score of 0.3 instead of 0.8 maintains more variants, mainly the rare SNPs (Figure 2.6), and removes unreliable imputed variants.

(ii) For different platforms such as sequences, the report of the aggregated imputation performance (e.g. with GLIMPSE) as a function of the MAFs as described in step 10 should define the threshold to adopt more specifically (as per Figure 2.6).

11| Optional step: Comparing runs with and without stringent pre-filtration (Figure 3)

This step is optional to compare the results of the dataset post-imputation with or without conservative QC at pre-filtration within the MAFs bins as previously described⁹⁵. The non-parametric Wilcoxon paired test is performed in each frequency class to compare the MAFs (Table 1) and information-IMPUTE2 scores (Table 2). Unless mentioned as one-sided, the two-sided Wilcoxon test in R is performed. Spearman’s correlations are also evaluated (Figures 2.4, 2.5). The r2_type0 accuracy metric within the high bins of MAFs was significantly increased (p=1E-06). The information scores are also significantly improved in all bins of MAFs (Table 2).

Overall, to palliate the discordance in the labelling of the SNPs, the chip, the reference genome for imputation and the control MAF database would need to be from the same human genome build version.

CLUSTERING RARE VARIANTS

Problem

It might be difficult to call rare variants in GBS or to cluster without ambiguity rare variants from microarray chips during the automatic creation of the calls in GenomeStudio.

Possible reason

Low numbers of rare variants are present in cohorts, especially with less than 1,000 individuals.

(i) Solving the problem with filtrations

Any variant with less than one heterozygous individual will be excluded by pre-filtration and later in the first step of post-filtration based on MAF-DH. In the second step at post-filtration, the only few remaining imputed SNPs with an information score < 0.3 would be further removed.

(ii) Solving the problem with genotyping

In larger sample sets, there are more individuals to create clusters of rare SNPs, e.g., MAF 0.1% would represent for 10,000 individuals at least 19 heterozygous individuals and 1 homozygous individual⁹⁶.

(iii) Solving the problem with a diversity panel

As a further approach since panels of rare variants would require more individuals, it will be useful to identify a) some individuals in existing panels carrying rare variants, e.g. by using different populations (e.g., HGDP, TopMed) and b) to genotype a diversity panel by microarray chip to include more rare variants.

Common variants in one population could become rare in another population of the diversity panel^115,123,124. Thus, to help with the clustering of rare variant calls, a) individual sets carrying rare SNPs and b) a diversity panel (the .idat files and the sample sheets) could be loaded into the Illumina GenomeStudio to overlap with the variants from each project studied by genotyping array at a given centre. This would increase the number of rare allele calls to create the clusters (AA, AB, BB) manually or automatically^118,125 by microarray when not enough individuals are available in a given study.

(iv) Improving calls for rare variants in sequencing

The calls for variants in low-pass sequencing or rare variants in sequencing can be improved with joint calling of additional samples and with the use of both strands^104,107,126. Although to recover AOH (absence of heterozygosity), higher depth coverage (5 to 8x up to 30x) would be necessary, as the lowest depth might not be efficient¹⁰⁶.

(v) Solving the problem with a control database

As a further control of accuracy, upload a reference SNP in an SQL database or query an online database with similar numbers of subjects and ethnicities to the dataset to compare the MAF and alleles at each variant (e.g. Figure 2.3).

(vi) Alternative Genotyping methods for control

The Sanger method traditionally used to sequence individuals on a P³²gel following PCR was able to identify, without ambiguity, the homozygous and heterozygous alleles. This approach was later replaced by automated methods¹²⁷. A restriction site either present or included by PCR served to confirm the locus of interest to genotype rare markers without ambiguity¹⁹. Relevant markers or rare variants could therefore be accurately Sanger sequenced and genotyped by restriction digest when no options to fully characterise the SNPs other than imputation are available.

(vii) Flags for downstream analyses

Variants showing a departure from HWE and those with INFO scores between 0.3 and 0.8 can be flagged for downstream analyses for further investigation.

The time taken is similar to that of a standard QC for GWAS (pre-filtration) with an additional time to compute the MAFs of the sample set. There is a further post-filtration step with the MAFs post-imputation, which would also slightly increase the time of QC. The procedure adds 15 minutes compared to a standard imputation QC. The choice for the MAF control database to use in step 6 will need to be adressed before the start of the project, as it depends on the study. The time of preparation for the MAF control database is not included in the protocol, as some public databases are available in different formats.

There is no consensus for imputation filtration threshold reliability to apply to downstream analyses in search of new loci. We find that the two-step post-filtration approach based on 1) the MAFs and 2) the quality imputation score could serve to reach consensus. This should avoid a conservative method with post-filtrations at INFO scores of 0.8 that are too stringent or one-step post-filtration scores of 0.3 that do not maintain enough confidence during the post-filtration of the variants¹²⁸. Including markers >= MAF of the sample set (MAF-DS) also limits the exclusion of markers at pre-filtration and ameliorates the performance of the imputation (Figure 3). The MAFs <0.01 represent the majority of the imputed variants (68%) (Table 2). Downstream imputation parameters are not affected by the procedure. The 2Mb region of the chromosome 20, a surrogate of the human genome, showed a global variant distribution similar to that of the entire genome⁹⁵. For larger sample sets, the confidence for the low MAFs will improve, as more individuals will be represented at rare MAFs in the dataset.

Therefore, when genotyping results are available in GWAS databases for pre-filtration, consider using the MAF of the sample set (MAF-DS) to increase the number of variants to impute and to improve the post-imputation information metric scores. The MAF computed for post-filtration (MAF-DH) also warrants more than one heterozygous individual (1.5) and just 1% of variants with INFO < 0.3 to eliminate and ensures a smaller loss of variants, including rare SNPs.

Imputed data from the early 1000GP panel could therefore be re-analysed using our 2-step procedure to incorporate markers with quality scores between 0.3-0.8 initially excluded from GWAS imputation. In return, it would increase the number of SNPs for association studies and shared variants for meta-analyses. The procedure increases the number of variants by 18% compared to conservative methods with one-step post-filtration at an INFO score of 0.8. Thus, it increases the gain of rare variants imputed that could be important in disease risk discovery.

In sequencing, the minimum allele frequency (e.g. MAF-DS) will need to be represented in at least one of the reads. Increasing the number of individuals in joint variant calling from sequences can improve the sensitivity of called genotypes to impute with the use of both strands^107,126 for rare variants. The imputation should further improve low MAFs and low-pass sequencing as the number of subjects is increased in the reference panels and the methodologies for phasing and imputation are developed (BEAGLE v4.1⁸⁴, GENEIMP v1.3⁹⁰, GLIMPSE⁹¹, LOIMPUTE v0.1.1⁹², QUILT⁹³ or STITCH v1.6.2⁹⁴).

Additionally, a mimic depth coverage of 1x low-pass sequencing for 503 EUR from the 2,504 1kGP 30x would show an average empirical r² of 0.8 (r² imputed vs. 1kGP 30x depth coverage) after imputation with GLIMPSE^68,91 if we apply the MAF-DH of these subjects at 0.0015.

In this instance, the first step of post-filtration based on the MAF-DH would produce a reliable imputation; in the second step, the threshold based on the recommended quality would be applied. However, the extrapolation remains arbitrary since the reported r² for the variants from the simulated data (503 from 2,504 1kGP 30x) by Rubinacci et al. 2021 were also represented against gnomAD v3 aggregated MAFs from 76,156 other individuals⁶⁹. Furthermore, gnomaD v3 reached lower MAFs and MAF-DS due to the large increase in the number of samples (76,156) compared to the study sample (503). In the literature, the imputation of genotype performance below 1x dropped as the depth coverage decreased, and the number of samples in the reference panel was reduced⁹¹.

1. Lathrop, G. M., Lalouel, J. M., Julier, C. & Ott, J. Strategies for multilocus linkage analysis in humans. Proceedings of the National Academy of Sciences of the United States of America 81, 3443-3446, doi:10.1073/pnas.81.11.3443 (1984).

2. Petrukhin, K. E. et al.A microsatellite genetic linkage map of human chromosome 13. Genomics 15, 76-85, doi:10.1006/geno.1993.1012 (1993).

3. Bycroft, C. et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 562, 203-209, doi:10.1038/s41586-018-0579-z (2018).

4. Cohen, D., Chumakov, I. & Weissenbach, J. A first-generation physical map of the human genome. Nature 366, 698-701, doi:10.1038/366698a0 (1993).

5. Hudson, T. J. et al. An STS-based map of the human genome. Science (New York, N.Y.) 270, 1945-1954, doi:10.1126/science.270.5244.1945 (1995).

6. Matthews, D. et al.Evidence that a locus for familial psoriasis maps to chromosome 4q. Nat Genet 14, 231-233, doi:10.1038/ng1096-231 (1996).

7. Weber, J. L. & May, P. E. Abundant class of human DNA polymorphisms which can be typed using the polymerase chain reaction. American journal of human genetics 44, 388-396 (1989).

8. Klebe, S. et al.Spastic paraplegia 5: Locus refinement, candidate gene analysis and clinical description. American journal of medical genetics. Part B, Neuropsychiatric genetics : the official publication of the International Society of Psychiatric Genetics 144b, 854-861, doi:10.1002/ajmg.b.30518 (2007).

9. Senee, V. et al.Mutations in GLIS3 are responsible for a rare syndrome with neonatal diabetes mellitus and congenital hypothyroidism. Nat Genet 38, 682-687, doi:10.1038/ng1802 (2006).

10. Deloukas, P. et al. A physical map of 30,000 human genes. Science (New York, N.Y.) 282, 744-746, doi:10.1126/science.282.5389.744 (1998).

11. Schuler, G. D. et al.A gene map of the human genome. Science (New York, N.Y.) 274, 540-546 (1996).

12. Zinovieva, E. et al.Comprehensive linkage and association analyses identify haplotype, near to the TNFSF15 gene, significantly associated with spondyloarthritis. PLoS Genet 5, e1000528, doi:10.1371/journal.pgen.1000528 (2009).

13. Parkinson, J. et al.Variation at the IRF2 gene and susceptibility to psoriasis in chromosome 4q-linked families. The Journal of investigative dermatology 122, 640-643, doi:10.1046/j.0022-202X.2004.22135.x (2004).

14. Willer, C. J., Li, Y. & Abecasis, G. R. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics (Oxford, England) 26, 2190-2191, doi:10.1093/bioinformatics/btq340 (2010).

15. International HapMap, C. The International HapMap Project. Nature 426, 789-796, doi:10.1038/nature02168 (2003).

16. International HapMap, C. A haplotype map of the human genome. Nature 437, 1299-1320, doi:10.1038/nature04226 (2005).

17. International HapMap, C.et al. A second generation human haplotype map of over 3.1 million SNPs. Nature 449, 851-861, doi:10.1038/nature06258 (2007).

18. Charon, C. M. & Dolphin, C. T. Sequence variation in the human flavin-containing monooxygenase 3 gene (FMO3): identification by denaturing-high performance liquid chromatography (DHPLC). Biochemical Society transactions 28, A435-A435, doi:https://doi.org/10.1042/bst028a435. From the 18th International Congress of Biochemistry and Molecular biology. Beyond the genome. (IUBMB International Union of Biochemistry and Molecular Biology/FEBS Congress : Program : Toxicology and xenobiochemistry. N°1588, July 20). July 16-20. Birmingham, UK. (2000).

19. Charon, C., Steventon, G. & Dolphin, C. New genetic variability of flavin containing monooxygenase 3 and its functional characterization. 54th Annual Meeting of the Amercian Society of Human Genetics. (ASHG : Program Nr 1526, October 29). October 26-30. Toronto, Ontario, Canada. Poster abstract n° 1526/F4: https://www.ashg.org/wp-content/uploads/2019/10/2004-poster-abstracts.pdf. , Page 1240 (2004).

20. Consortium., G. P. et al.An integrated map of genetic variation from 1,092 human genomes. Nature 491, 56-65, doi:10.1038/nature11632 (2012).

21. Bentley, D. R. et al.Accurate whole human genome sequencing using reversible terminator chemistry. Nature 456, 53-59, doi:10.1038/nature07517 (2008).

22. Rabbee, N. & Speed, T. P. A genotype calling algorithm for affymetrix SNP arrays. Bioinformatics (Oxford, England) 22, 7-12, doi:10.1093/bioinformatics/bti741 (2006).

23. Huc-Chabrolle, M. et al.Xq27 FRAXA locus is a strong candidate for dyslexia: evidence from a genome-wide scan in French families. Behavior genetics 43, 132-140, doi:10.1007/s10519-012-9575-5 (2013).

24. Caburet, S. et al.Genome-wide linkage in a highly consanguineous pedigree reveals two novel loci on chromosome 7 for non-syndromic familial Premature Ovarian Failure. PLoS One 7, e33412, doi:10.1371/journal.pone.0033412 (2012).

25. Schaid, D. J., Chen, W. & Larson, N. B. From genome-wide associations to candidate causal variants by statistical fine-mapping. Nat Rev Genet 19, 491-504, doi:10.1038/s41576-018-0016-z (2018).

26. Stocks, T. et al.TFAP2B -dietary protein and glycemic index interactions and weight maintenance after weight loss in the DiOGenes trial. Hum Hered 75, 213-219, doi:10.1159/000353591 (2013).

27. Matone, A. et al.Network Analysis of Metabolite GWAS Hits: Implication of CPS1 and the Urea Cycle in Weight Maintenance. PLoS One 11, e0150495, doi:10.1371/journal.pone.0150495 (2016).

28. Costantino, F. et al.Whole-genome single nucleotide polymorphism-based linkage analysis in spondyloarthritis multiplex families reveals a new susceptibility locus in 13q13. Annals of the rheumatic diseases 75, 1380-1385, doi:10.1136/annrheumdis-2015-207720 (2016).

29. Lambert, J. C. et al.Genome-wide association study identifies variants at CLU and CR1 associated with Alzheimer's disease. Nat Genet 41, 1094-1099, doi:10.1038/ng.439 (2009).

30. Marchini, J., Howie, B., Myers, S., McVean, G. & Donnelly, P. A new multipoint method for genome-wide association studies by imputation of genotypes. Nat Genet 39, 906-913, doi:10.1038/ng2088 (2007).

31. Li, Y., Willer, C., Sanna, S. & Abecasis, G. Genotype imputation. Annu Rev Genomics Hum Genet 10, 387-406, doi:10.1146/annurev.genom.9.081307.164242 (2009).

32. Spencer, C. C., Su, Z., Donnelly, P. & Marchini, J. Designing genome-wide association studies: sample size, power, imputation, and the choice of genotyping chip. PLoS Genet 5, e1000477, doi:10.1371/journal.pgen.1000477 (2009).

33. Anderson, C. A. et al.Evaluating the effects of imputation on the power, coverage, and cost efficiency of genome-wide SNP platforms. American journal of human genetics 83, 112-119, doi:10.1016/j.ajhg.2008.06.008 (2008).

34. Hao, K., Chudin, E., McElwee, J. & Schadt, E. E. Accuracy of genome-wide imputation of untyped markers and impacts on statistical power for association studies. BMC genetics 10, 27, doi:10.1186/1471-2156-10-27 (2009).

35. Darabi, H. et al. Fine scale mapping of the 17q22 breast cancer locus using dense SNPs, genotyped within the Collaborative Oncological Gene-Environment Study (COGs). Scientific reports 6, 32512, doi:10.1038/srep32512 (2016).

36. Orho-Melander, M. et al.Common missense variant in the glucokinase regulatory protein gene is associated with increased plasma triglyceride and C-reactive protein but lower fasting glucose concentrations. Diabetes 57, 3112-3121, doi:10.2337/db08-0516 (2008).

37. Vigorito, E. et al.Fine-Scale Mapping at 9p22.2 Identifies Candidate Causal Variants That Modify Ovarian Cancer Risk in BRCA1 and BRCA2 Mutation Carriers. PLoS One 11, e0158801, doi:10.1371/journal.pone.0158801 (2016).

38. Jonsson, T. et al. A mutation in APP protects against Alzheimer's disease and age-related cognitive decline. Nature 488, 96-99, doi:10.1038/nature11283 (2012).

39. Jonsson, T. et al.Variant of TREM2 associated with the risk of Alzheimer's disease. The New England journal of medicine 368, 107-116, doi:10.1056/NEJMoa1211103 (2013).

40. Lin, D. Y. & Zeng, D. Meta-analysis of genome-wide association studies: no efficiency gain in using individual participant data. Genetic epidemiology 34, 60-66, doi:10.1002/gepi.20435 (2010).

41. Willer, C. J. et al.Newly identified loci that influence lipid concentrations and risk of coronary artery disease. Nat Genet 40, 161-169, doi:10.1038/ng.76 (2008).

42. Collins, F. S. Reengineering Translational Science: The Time Is Right. Science Translational Medicine 3, 90cm17-90cm17, doi:doi:10.1126/scitranslmed.3002747 (2011).

43. Wain, L. V. et al.Genome-wide association analyses for lung function and chronic obstructive pulmonary disease identify new loci and potential druggable targets. Nat Genet 49, 416-425, doi:10.1038/ng.3787 (2017).

44. Lau, A. & So, H. C. Turning genome-wide association study findings into opportunities for drug repositioning. Computational and structural biotechnology journal 18, 1639-1650, doi:10.1016/j.csbj.2020.06.015 (2020).

45. Jhamb, D., Magid-Slav, M., Hurle, M. R. & Agarwal, P. Pathway analysis of GWAS loci identifies novel drug targets and repurposing opportunities. Drug discovery today 24, 1232-1236, doi:10.1016/j.drudis.2019.03.024 (2019).

46. Nagel, M. et al.Meta-analysis of genome-wide association studies for neuroticism in 449,484 individuals identifies novel genetic loci and pathways. Nat Genet 50, 920-927, doi:10.1038/s41588-018-0151-7 (2018).

47. Pericak-Vance, M. A. et al. Linkage studies in familial Alzheimer disease: evidence for chromosome 19 linkage. American journal of human genetics 48, 1034-1050 (1991).

48. Corder, E. H. et al.Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer's disease in late onset families. Science (New York, N.Y.) 261, 921-923, doi:10.1126/science.8346443 (1993).

49. Strittmatter, W. J. Apolipoprotein E and Alzheimer's disease: signal transduction mechanisms. Biochemical Society symposium, 101-109, doi:10.1042/bss0670101 (2001).

50. Saunders, A. M. et al.Association of apolipoprotein E allele epsilon 4 with late-onset familial and sporadic Alzheimer's disease. Neurology 43, 1467-1472, doi:10.1212/wnl.43.8.1467 (1993).

51. Farrer, L. A. et al.Effects of age, sex, and ethnicity on the association between apolipoprotein E genotype and Alzheimer disease. A meta-analysis. APOE and Alzheimer Disease Meta Analysis Consortium. Jama 278, 1349-1356 (1997).

52. Taubes, A. et al.Experimental and real-world evidence supporting the computational repurposing of bumetanide for APOE4-related Alzheimer’s disease. Nature Aging 1, 932-947, doi:10.1038/s43587-021-00122-7 (2021).

53. Jansen, I. E. et al.Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer's disease risk. Nat Genet 51, 404-413, doi:10.1038/s41588-018-0311-9 (2019).

54. Reay, W. R. & Cairns, M. J. Advancing the use of genome-wide association studies for drug repurposing. Nature Reviews Genetics 22, 658-671, doi:10.1038/s41576-021-00387-z (2021).

55. Dilthey, A. et al.Multi-population classical HLA type imputation. PLoS Comput Biol 9, e1002877, doi:10.1371/journal.pcbi.1002877 (2013).

56. Mavaddat, N. et al.Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes. American journal of human genetics 104, 21-34, doi:10.1016/j.ajhg.2018.11.002 (2019).

57. Frazer, K. A. et al. A second generation human haplotype map of over 3.1 million SNPs. Nature 449, 851-861, doi:10.1038/nature06258 (2007).

58. Huang, L. et al.Genotype-imputation accuracy across worldwide human populations. American journal of human genetics 84, 235-250, doi:10.1016/j.ajhg.2009.01.013 (2009).

59. Genomes Project, C. et al.A global reference for human genetic variation. Nature 526, 68-74, doi:10.1038/nature15393 (2015).

60. Sudmant, P. H. et al.An integrated map of structural variation in 2,504 human genomes. Nature 526, 75-81, doi:10.1038/nature15394 (2015).

61. Hehir-Kwa, J. Y. et al.A high-quality human reference panel reveals the complexity and distribution of genomic structural variants. Nat Commun 7, 12989, doi:10.1038/ncomms12989 (2016).

62. McCarthy, S. et al. A reference panel of 64,976 haplotypes for genotype imputation. Nat Genet 48, 1279-1283, doi:10.1038/ng.3643 (2016).

63. Delaneau, O., Marchini, J. & Consortium., T. G. P. Integrating sequence and array data to create an improved 1000 Genomes Project haplotype reference panel. Nat Commun 5, 3934, doi:10.1038/ncomms4934 (2014).

64. Iotchkova, V. et al.Discovery and refinement of genetic loci associated with cardiometabolic risk using dense imputation maps. Nat Genet 48, 1303-1312, doi:10.1038/ng.3668 (2016).

65. Huang, J. et al.Improved imputation of low-frequency and rare variants using the UK10K haplotype reference panel. Nat Commun 6, 8111, doi:10.1038/ncomms9111 (2015).

66. Vernon Smith, A. NHLBI TOPMed program : How to use WGS and other multi-omics data on over 140K genetically diverse and deeply phenotyped participants to advance your science : Genomic Variation in TOPMed. 70th Annual Meeting of the Amercian Society of Human Genetics (ASHG : NHLBI TOPMed program : ancillary session, October 26). October 27-30, Virtual meeting. Presentation : https://topmed.nhlbi.nih.gov/sites/default/files/Genomic%20Variation%20in%20TOPMed%20-%20IRC.pdf, https://www.abstractsonline.com/pp8/#!/9070/session/548. (2020).

67. Taliun, D. et al.Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature 590, 290-299, doi:10.1038/s41586-021-03205-y (2021).

68. Byrska-Bishop, M. et al.High-coverage whole-genome sequencing of the expanded 1000 Genomes Project cohort including 602 trios. Cell 185, 3426-3440.e3419, doi:10.1016/j.cell.2022.08.004 (2022).

69. Gudmundsson, S. et al.Variant interpretation using population databases: Lessons from gnomAD. Human Mutation 43, 1012-1030, doi:https://doi.org/10.1002/humu.24309 (2022).

70. Karczewski, K. J. et al.The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434-443, doi:10.1038/s41586-020-2308-7 (2020).

71. Lek, M. et al.Analysis of protein-coding genetic variation in 60,706 humans. Nature 536, 285-291, doi:10.1038/nature19057 (2016).

72. Vergara, C. et al.Genotype imputation performance of three reference panels using African ancestry individuals. Human genetics 137, 281-292, doi:10.1007/s00439-018-1881-4 (2018).

73. Gurdasani, D. et al.The African Genome Variation Project shapes medical genetics in Africa. Nature 517, 327-332, doi:10.1038/nature13997 (2015).

74. Rotimi, C. et al.Research capacity. Enabling the genomic revolution in Africa. Science (New York, N.Y.) 344, 1346-1348, doi:10.1126/science.1251546 (2014).

75. Mathias, R. A. et al.A continuum of admixture in the Western Hemisphere revealed by the African Diaspora genome. Nat Commun 7, 12522, doi:10.1038/ncomms12522 (2016).

76. Schurz, H. et al.Evaluating the Accuracy of Imputation Methods in a Five-Way Admixed Population. Frontiers in genetics 10, 34, doi:10.3389/fgene.2019.00034 (2019).

77. Regan, E. A. et al.Genetic epidemiology of COPD (COPDGene) study design. Copd 7, 32-43, doi:10.3109/15412550903499522 (2010).

78. Browning, S. R. & Browning, B. L. Rapid and accurate haplotype phasing and missing-data inference for whole-genome association studies by use of localized haplotype clustering. American journal of human genetics 81, 1084-1097, doi:10.1086/521987 (2007).

79. Browning, S. R. & Browning, B. L. Haplotype phasing: existing methods and new developments. Nat Rev Genet 12, 703-714, doi:10.1038/nrg3054 (2011).

80. Browning, B. L. & Browning, S. R. A unified approach to genotype imputation and haplotype-phase inference for large data sets of trios and unrelated individuals. American journal of human genetics 84, 210-223, doi:10.1016/j.ajhg.2009.01.005 (2009).

81. Li, N. & Stephens, M. Modeling linkage disequilibrium and identifying recombination hotspots using single-nucleotide polymorphism data. Genetics 165, 2213-2233 (2003).

82. Howie, B. N., Donnelly, P. & Marchini, J. A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genet 5, e1000529, doi:10.1371/journal.pgen.1000529 (2009).

83. Li, Y., Willer, C. J., Ding, J., Scheet, P. & Abecasis, G. R. MaCH: using sequence and genotype data to estimate haplotypes and unobserved genotypes. Genetic epidemiology 34, 816-834, doi:10.1002/gepi.20533 (2010).

84. Browning, B. L. & Browning, S. R. Genotype Imputation with Millions of Reference Samples. American journal of human genetics 98, 116-126, doi:10.1016/j.ajhg.2015.11.020 (2016).

85. Marchini, J. & Howie, B. Genotype imputation for genome-wide association studies. Nat Rev Genet 11, 499-511, doi:10.1038/nrg2796 (2010).

86. Delaneau, O., Zagury, J. F. & Marchini, J. Improved whole-chromosome phasing for disease and population genetic studies. Nat Methods 10, 5-6, doi:10.1038/nmeth.2307 (2013).

87. Delaneau, O., Zagury, J. F., Robinson, M. R., Marchini, J. L. & Dermitzakis, E. T. Accurate, scalable and integrative haplotype estimation. Nat Commun 10, 5436, doi:10.1038/s41467-019-13225-y (2019).

88. Rubinacci, S., Delaneau, O. & Marchini, J. Genotype imputation using the Positional Burrows Wheeler Transform. PLoS Genet 16, e1009049, doi:10.1371/journal.pgen.1009049 (2020).

89. Durbin, R. Efficient haplotype matching and storage using the positional Burrows-Wheeler transform (PBWT). Bioinformatics (Oxford, England) 30, 1266-1272, doi:10.1093/bioinformatics/btu014 (2014).

90. Spiliopoulou, A., Colombo, M., Orchard, P., Agakov, F. & McKeigue, P. GeneImp: Fast Imputation to Large Reference Panels Using Genotype Likelihoods from Ultralow Coverage Sequencing. Genetics 206, 91-104, doi:10.1534/genetics.117.200063 (2017).

91. Rubinacci, S., Ribeiro, D. M., Hofmeister, R. J. & Delaneau, O. Efficient phasing and imputation of low-coverage sequencing data using large reference panels. Nat Genet 53, 120-126, doi:10.1038/s41588-020-00756-0 (2021).

92. Wasik, K. et al.Comparing low-pass sequencing and genotyping for trait mapping in pharmacogenetics. BMC Genomics 22, 197, doi:10.1186/s12864-021-07508-2 (2021).

93. Davies, R. W. et al.Rapid genotype imputation from sequence with reference panels. Nat Genet 53, 1104-1111, doi:10.1038/s41588-021-00877-0 (2021).

94. Davies, R. W., Flint, J., Myers, S. & Mott, R. Rapid genotype imputation from sequence without reference panels. Nature Genetics 48, 965-969, doi:10.1038/ng.3594 (2016).

95. Charon, C., Allodji, R., Meyer, V. & Deleuze, J. F. Impact of pre- and post-variant filtration strategies on imputation. Scientific reports 11, 6214, doi:10.1038/s41598-021-85333-z (2021).

96. Coleman, J. R. et al.Quality control, imputation and analysis of genome-wide genotyping data from the Illumina HumanCoreExome microarray. Brief Funct Genomics 15, 298-304, doi:10.1093/bfgp/elv037 (2016).

97. Sanchez, M. P. et al.Short communication: Confirmation of candidate causative variants on milk composition and cheesemaking properties in Montbeliarde cows. J Dairy Sci 101, 10076-10081, doi:10.3168/jds.2018-14986 (2018).

98. Mancin, E., Sosa-Madrid, B. S., Blasco, A. & Ibáñez-Escriche, N. Genotype Imputation to Improve the Cost-Efficiency of Genomic Selection in Rabbits. Animals 11, 803 (2021).

99. Keeble-Gagnère, G. et al.Novel Design of Imputation-Enabled SNP Arrays for Breeding and Research Applications Supporting Multi-Species Hybridization. Frontiers in Plant Science 12, doi:10.3389/fpls.2021.756877 (2021).

100. Kim, M., Kim, J. H., Kim, K. & Kim, S. Cost-effective and accurate method of measuring fetal fraction using SNP imputation. Bioinformatics (Oxford, England) 34, 1086-1091, doi:10.1093/bioinformatics/btx728 (2018).

101. Das, S., Abecasis, G. R. & Browning, B. L. Genotype Imputation from Large Reference Panels. Annu Rev Genomics Hum Genet 19, 73-96, doi:10.1146/annurev-genom-083117-021602 (2018).

102. Browning, B. L., Zhou, Y. & Browning, S. R. A One-Penny Imputed Genome from Next-Generation Reference Panels. The American Journal of Human Genetics 103, 338-348, doi:https://doi.org/10.1016/j.ajhg.2018.07.015 (2018).

103. Tachmazidou, I. et al.Whole-Genome Sequencing Coupled to Imputation Discovers Genetic Signals for Anthropometric Traits. American journal of human genetics 100, 865-884, doi:10.1016/j.ajhg.2017.04.014 (2017).

104. Lou, R. N., Jacobs, A., Wilder, A. P. & Therkildsen, N. O. A beginner's guide to low-coverage whole genome sequencing for population genomics. Mol Ecol 30, 5966-5993, doi:10.1111/mec.16077 (2021).

105. Li, J. H., Mazur, C. A., Berisa, T. & Pickrell, J. K. Low-pass sequencing increases the power of GWAS and decreases measurement error of polygenic risk scores compared to genotyping arrays. Genome research 31, 529-537, doi:10.1101/gr.266486.120 (2021).

106. Chaubey, A. et al.Low-Pass Genome Sequencing: Validation and Diagnostic Utility from 409 Clinical Cases of Low-Pass Genome Sequencing for the Detection of Copy Number Variants to Replace Constitutional Microarray. J Mol Diagn 22, 823-840, doi:10.1016/j.jmoldx.2020.03.008 (2020).

107. DePristo, M. A. et al.A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nature Genetics 43, 491-498, doi:10.1038/ng.806 (2011).

108. Zheng, C., Boer, M. P. & van Eeuwijk, F. A. Accurate Genotype Imputation in Multiparental Populations from Low-Coverage Sequence. Genetics 210, 71-82, doi:10.1534/genetics.118.300885 (2018).

109. R : A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL - Bioconductor 3.2. R.version 3.2.4 in Rstudio 0.99.115 - 2016-03-10 (2016).

110. Purcell, S. et al.PLINK: a tool set for whole-genome association and population-based linkage analyses. American journal of human genetics 81, 559-575, doi:10.1086/519795 (2007).

111. Purcell, S. PLINK: Whole genome data analysis toolset (v 1.07). https://zzz.bwh.harvard.edu/plink/download.shtml (2015).

112. Delaneau, O., Marchini, J. & Zagury, J.-F. A linear complexity phasing method for thousands of genomes. Nature Methods 9, 179-181, doi:10.1038/nmeth.1785 (2012).

113. O'Connell, J. et al.Haplotype estimation for biobank-scale data sets. Nat Genet 48, 817-820, doi:10.1038/ng.3583 (2016).

114. Patterson, M. et al.WhatsHap: Weighted Haplotype Assembly for Future-Generation Sequencing Reads. Journal of computational biology : a journal of computational molecular cell biology 22, 498-509, doi:10.1089/cmb.2014.0157 (2015).

115. Howie, B., Marchini, J. & Stephens, M. Genotype imputation with thousands of genomes. G3 (Bethesda, Md.) 1, 457-470, doi:10.1534/g3.111.001198 (2011).

116. Wellcome Trust Case Control, C. Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 447, 661-678, doi:10.1038/nature05911 (2007).

117. Charon, C., Allodji, R. & Deleuze, J. F. Effects of filtrations on imputation in clusterised variants. 67th Annual Meeting of the Amercian Society of Human Genetics (ASHG : Program : Bioinformatics and Computational Approaches, October 18). October 17-21. Orlando, Florida, USA. Poster abstract 1444W : https://www.ashg.org/wp-content/uploads/2019/10/2017-Poster-Abstracts.pdf. , Page 537 (2017).

118. Guo, Y. et al.Illumina human exome genotyping array clustering and quality control. Nature Protocols 9, 2643-2662, doi:10.1038/nprot.2014.174 (2014).

119. Cavalli-Sforza, L. L. The Human Genome Diversity Project: past, present and future. Nature Reviews Genetics 6, 333-340, doi:10.1038/nrg1596 (2005).

120. Beck, T. F., Mullikin, J. C. & Biesecker, L. G. Systematic Evaluation of Sanger Validation of Next-Generation Sequencing Variants. Clinical chemistry 62, 647-654, doi:10.1373/clinchem.2015.249623 (2016).

121. Mu, W., Lu, H. M., Chen, J., Li, S. & Elliott, A. M. Sanger Confirmation Is Required to Achieve Optimal Sensitivity and Specificity in Next-Generation Sequencing Panel Testing. J Mol Diagn 18, 923-932, doi:10.1016/j.jmoldx.2016.07.006 (2016).

122. Price, A. L. et al.Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet 38, 904-909, doi:10.1038/ng1847 (2006).

123. Huang, G. H. & Tseng, Y. C. Genotype imputation accuracy with different reference panels in admixed populations. BMC Proc 8, S64, doi:10.1186/1753-6561-8-S1-S64 (2014).

124. Mathieson, I. & McVean, G. Demography and the age of rare variants. PLoS Genet 10, e1004528, doi:10.1371/journal.pgen.1004528 (2014).

125. Goldstein, J. I. et al.zCall: a rare variant caller for array-based genotyping: genetics and population analysis. Bioinformatics (Oxford, England) 28, 2543-2545, doi:10.1093/bioinformatics/bts479 (2012).

126. Koboldt, D. C. Best practices for variant calling in clinical sequencing. Genome Med 12, 91, doi:10.1186/s13073-020-00791-w (2020).

127. Heather, J. M. & Chain, B. The sequence of sequencers: The history of sequencing DNA. Genomics 107, 1-8, doi:10.1016/j.ygeno.2015.11.003 (2016).

128. Charon, C., Allodji, R., Deleuze, J. F. & Meyer, V. Evaluation de strategies de filtration de variants genetiques lors d'analyses d'imputation effectueées dans la recherche de facteurs de susceptibilite aux pathologies humaines. Scoop.it. Life Sciencdes Universite Paris-Saclay. 4th April 2020. (http://sco.lt/7avC08 : https://www.scoop.it/topic/life-sci-news-upsaclay/p/4124072256/2021/04/04/evaluation-de-strategies-de-filtration-de-variants-genetiques-lors-d-analyses-d-imputation-effectuees-dans-la-recherche-de-facteurs-de-susceptibilite-aux-pathologies-humaines) (2021).

We thank Steven McGinn for his assistance in English language editing, and Stephane Cretoni for his advice in the drawing of Figure 1. We also thank A. Leduc from the Bioinformatics Department at CNRGH (formerly CNG) for building the SQL database of NCBI build 37, Fernando Dos-Santos for confirming the full retrieval of data from the SQL database, and E Larsonneur with N. Wiart for guiding through the workflow management system at CNRGH.

The AJE issued on Novembrer 29, 2022 an editing certificate number 9D09-9485-F3BF-B9D6-F5F7 for proper English language after reviewing the language and overall style of the manuscript. Final editing was performed by the author.

Download PDF

Version 1

posted

You are reading this latest protocol version

Pre- and post-genotype filtration protocol to improve the variant imputation metrics with further quality control

Status:

Version 1

Abstract

Figures

Introduction

Equipment

Procedure

Troubleshooting

Time Taken

Anticipated Results

References

Acknowledgements

Associated Publications

Status:

Version 1

Privacy Policy

Terms of Service

Pre- and post-genotype filtration protocol to improve the variant imputation metrics with further quality control

Status:

Version 1

Abstract

Figures

Introduction

Equipment

Procedure

Troubleshooting

Time Taken

Anticipated Results

References

Acknowledgements

Associated Publications

Status:

Version 1

Privacy Policy

Terms of Service

Manage Cookie Preferences