Limits...
Low-depth genotyping-by-sequencing (GBS) in a bovine population: strategies to maximize the selection of high quality genotypes and the accuracy of imputation

View Article: PubMed Central - PubMed

ABSTRACT

Background: Genotyping-by-sequencing (GBS) has emerged as a powerful and cost-effective approach for discovering and genotyping single-nucleotide polymorphisms. The GBS technique was largely used in crop species where its low sequence coverage is not a drawback for calling genotypes because inbred lines are almost homozygous. In contrast, only a few studies used the GBS technique in animal populations (with sizeable heterozygosity rates) and many of those that have been published did not consider the quality of the genotypes produced by the bioinformatic pipelines. To improve the sequence coverage of the fragments, an alternative GBS preparation protocol that includes selective primers during the PCR amplification step has been recently proposed. In this study, we compared this modified protocol with the conventional two-enzyme GBS protocol. We also described various procedures to maximize the selection of high quality genotypes and to increase the accuracy of imputation.

Results: The in silico digestions of the bovine genome showed that the combination of PstI and MspI is more suitable for sequencing bovine GBS libraries than the use of single digestions with PstI or ApeKI. The sequencing output of the GBS libraries generated a total of 123,666 variants with the selective-primer approach and 272,103 variants with the conventional approach. Validating our data with genotypes obtained from mass spectrometry and Illumina’s bovine SNP50 array, we found that the genotypes produced by the conventional GBS method were concordant with those produced by these alternative genotyping methods, whereas the selective-primer method failed to call heterozygotes with confidence. Our results indicate that high accuracy in genotype calling (>97%) can be obtained using low read-depth thresholds (3 to 5 reads) provided that markers are simultaneously filtered for genotype quality scores. We also show that factors such as the minimum call rate and the minor allele frequency positively influence the accuracy of imputation of missing GBS data. The highest accuracies (around 85%) of imputed GBS markers were obtained with the FIMPUTE program when GBS and SNP50 array genotypes were combined (80,190 to 100,297 markers) before imputation.

Conclusions: We discovered that the conventional two-enzyme GBS protocol could produce a large number of high-quality genotypes provided that appropriate filtration criteria were used. In contrast, the selective-primer approach resulted in a substantial proportion of miscalled genotypes and should be avoided for livestock genotyping studies. Overall, our study demonstrates that carefully adjusting the different filtering parameters applied to the GBS data is critical to maximize the selection of high quality genotypes and to increase the accuracy of imputation of missing data. The strategies and results presented here provide a framework to maximize the output of the GBS technique in animal populations and qualified the PstI/MspI GBS assay as a low-cost high-density genotyping platform. The conclusions reported here regarding read-depth and genotype quality filtering could benefit many GBS applications, notably genome-wide association studies, where there is a need to increase the density of markers genotyped across the target population while preserving the quality of genotypes.

No MeSH data available.


Estimated accuracies of GBS genotypes after the first quality control step. Accuracy was estimated using GBS variants that were also found on the SNP50 array. The GBS genotypes were filtered according to their associated read-depth (RD) coverage without consideration of the genotype quality (GQ) or using a GQ score threshold of 20. At each shared locus, the concordance of the GBS and the SNP50 genotypes was assessed, assuming that those reported by the SNP50 array represented true genotypes
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC5382419&req=5

Fig5: Estimated accuracies of GBS genotypes after the first quality control step. Accuracy was estimated using GBS variants that were also found on the SNP50 array. The GBS genotypes were filtered according to their associated read-depth (RD) coverage without consideration of the genotype quality (GQ) or using a GQ score threshold of 20. At each shared locus, the concordance of the GBS and the SNP50 genotypes was assessed, assuming that those reported by the SNP50 array represented true genotypes

Mentions: Working henceforth with GBS genotypes produced by the conventional-primer approach, we were initially interested in determining the minimum number of reads required to call a genotype with confidence. We also examined how filtering simultaneously for GQ could affect the accuracy of genotypes. As shown in Fig. 5, we found that the minimum number of reads required to call a genotype had a great impact on the number of variants that passed this first filtering step. Again, this is explained by the above-mentioned distribution of read-depth values associated with genotypes (Fig. 4). Interestingly, when datasets were not filtered according to GQ, the accuracy substantially decreased, but this gap was reduced when the minimum read-depth threshold was increased (Fig. 5). For example, genotypes in dataset 1 (Table 3) had an overall estimated accuracy of only 92.8% in comparison with those in dataset 7 (97.7%). To explain this result, one must realise that many variants included in this dataset were called using only three reads and that the probability of miscalling a heterozygote using three reads is high, at 1 in 4. In contrast, simultaneous filtering for GQ consistently increases the genotype accuracy, because genotype calls having a GQ score greater than 20 (encoded as a Phred score) have a greater than 99% probability of being real.Fig. 5


Low-depth genotyping-by-sequencing (GBS) in a bovine population: strategies to maximize the selection of high quality genotypes and the accuracy of imputation
Estimated accuracies of GBS genotypes after the first quality control step. Accuracy was estimated using GBS variants that were also found on the SNP50 array. The GBS genotypes were filtered according to their associated read-depth (RD) coverage without consideration of the genotype quality (GQ) or using a GQ score threshold of 20. At each shared locus, the concordance of the GBS and the SNP50 genotypes was assessed, assuming that those reported by the SNP50 array represented true genotypes
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC5382419&req=5

Fig5: Estimated accuracies of GBS genotypes after the first quality control step. Accuracy was estimated using GBS variants that were also found on the SNP50 array. The GBS genotypes were filtered according to their associated read-depth (RD) coverage without consideration of the genotype quality (GQ) or using a GQ score threshold of 20. At each shared locus, the concordance of the GBS and the SNP50 genotypes was assessed, assuming that those reported by the SNP50 array represented true genotypes
Mentions: Working henceforth with GBS genotypes produced by the conventional-primer approach, we were initially interested in determining the minimum number of reads required to call a genotype with confidence. We also examined how filtering simultaneously for GQ could affect the accuracy of genotypes. As shown in Fig. 5, we found that the minimum number of reads required to call a genotype had a great impact on the number of variants that passed this first filtering step. Again, this is explained by the above-mentioned distribution of read-depth values associated with genotypes (Fig. 4). Interestingly, when datasets were not filtered according to GQ, the accuracy substantially decreased, but this gap was reduced when the minimum read-depth threshold was increased (Fig. 5). For example, genotypes in dataset 1 (Table 3) had an overall estimated accuracy of only 92.8% in comparison with those in dataset 7 (97.7%). To explain this result, one must realise that many variants included in this dataset were called using only three reads and that the probability of miscalling a heterozygote using three reads is high, at 1 in 4. In contrast, simultaneous filtering for GQ consistently increases the genotype accuracy, because genotype calls having a GQ score greater than 20 (encoded as a Phred score) have a greater than 99% probability of being real.Fig. 5

View Article: PubMed Central - PubMed

ABSTRACT

Background: Genotyping-by-sequencing (GBS) has emerged as a powerful and cost-effective approach for discovering and genotyping single-nucleotide polymorphisms. The GBS technique was largely used in crop species where its low sequence coverage is not a drawback for calling genotypes because inbred lines are almost homozygous. In contrast, only a few studies used the GBS technique in animal populations (with sizeable heterozygosity rates) and many of those that have been published did not consider the quality of the genotypes produced by the bioinformatic pipelines. To improve the sequence coverage of the fragments, an alternative GBS preparation protocol that includes selective primers during the PCR amplification step has been recently proposed. In this study, we compared this modified protocol with the conventional two-enzyme GBS protocol. We also described various procedures to maximize the selection of high quality genotypes and to increase the accuracy of imputation.

Results: The in silico digestions of the bovine genome showed that the combination of PstI and MspI is more suitable for sequencing bovine GBS libraries than the use of single digestions with PstI or ApeKI. The sequencing output of the GBS libraries generated a total of 123,666 variants with the selective-primer approach and 272,103 variants with the conventional approach. Validating our data with genotypes obtained from mass spectrometry and Illumina’s bovine SNP50 array, we found that the genotypes produced by the conventional GBS method were concordant with those produced by these alternative genotyping methods, whereas the selective-primer method failed to call heterozygotes with confidence. Our results indicate that high accuracy in genotype calling (>97%) can be obtained using low read-depth thresholds (3 to 5 reads) provided that markers are simultaneously filtered for genotype quality scores. We also show that factors such as the minimum call rate and the minor allele frequency positively influence the accuracy of imputation of missing GBS data. The highest accuracies (around 85%) of imputed GBS markers were obtained with the FIMPUTE program when GBS and SNP50 array genotypes were combined (80,190 to 100,297 markers) before imputation.

Conclusions: We discovered that the conventional two-enzyme GBS protocol could produce a large number of high-quality genotypes provided that appropriate filtration criteria were used. In contrast, the selective-primer approach resulted in a substantial proportion of miscalled genotypes and should be avoided for livestock genotyping studies. Overall, our study demonstrates that carefully adjusting the different filtering parameters applied to the GBS data is critical to maximize the selection of high quality genotypes and to increase the accuracy of imputation of missing data. The strategies and results presented here provide a framework to maximize the output of the GBS technique in animal populations and qualified the PstI/MspI GBS assay as a low-cost high-density genotyping platform. The conclusions reported here regarding read-depth and genotype quality filtering could benefit many GBS applications, notably genome-wide association studies, where there is a need to increase the density of markers genotyped across the target population while preserving the quality of genotypes.

No MeSH data available.