Turning Vice into Virtue: Using Batch-Effects to Detect Errors in Large Genomic Data Sets

Abstract

It is often unavoidable to combine data from different sequencing centers or sequencing platforms when compiling data sets with a large number of individuals. However, the different data are likely to contain specific systematic errors that will appear as SNPs. Here, we devise a method to detect systematic errors in combined data sets. To measure quality differences between individual genomes, we study pairs of variants that reside on different chromosomes and co-occur in individuals. The abundance of these pairs of variants in different genomes is then used to detect systematic errors due to batch effects. Applying our method to the 1000 Genomes data set, we find that coding regions are enriched for errors, where ∼1% of the higher frequency variants are predicted to be erroneous, whereas errors outside of coding regions are much rarer (<0.001%). As expected, predicted errors are found less often than other variants in a data set that was generated with a different sequencing technology, indicating that many of the candidates are indeed errors. However, predicted 1000 Genomes errors are also found in other large data sets; our observation is thus not specific to the 1000 Genomes data set. Our results show that batch effects can be turned into a virtue by using the resulting variation in large scale data sets to detect systematic errors.

The presence and impact of reference bias on population genomic studies of prehistoric human populations

“…Differences in sequencing strategies (e.g. read length) and bioinformatic processing have been shown to generate batch effects and dramatically affect downstream analyses [56–59]. Another well known bias in population genetics is ascertainment bias which arises when the studied variants were ascertained in selected populations only, and can substantially impact measurements of heterozygosity and related methods [60].…”

Negative linkage disequilibrium between amino acid changing variants reveals interference among deleterious mutations in the human genome

“…However, we see qualitatively similar patterns in 4 of the 5 populations from the 1KGP. Additionally, previous work has suggested that low-frequency errors that are the consequence of batch effects tend to be in positive LD with each other [41] . Also, variants that are identified as error candidates are more likely to be NS variants [41] .…”

“…Additionally, previous work has suggested that low-frequency errors that are the consequence of batch effects tend to be in positive LD with each other [41] . Also, variants that are identified as error candidates are more likely to be NS variants [41] . With this in mind, we suspect that sequencing errors would bias our analysis of pairs of low frequency variants annotated as being NS towards being more often in positive LD than in negative LD.…”

Legacy Data Confounds Genomics Studies

“…Mafessoni et al recently identified batch effects in the 1kGP by looking for individuals with excess LD among distant variants. (Mafessoni et al, 2018). Here we point out how these and additional unreported batch effects in the early phases of the 1kGP lead to incorrect genetic conclusions through population genetic analyses and spurious GWAS associations as a result of imputation using the 1kGP as a reference.…”

“…A recent publication by Mafessoni et al also identified a batch effect in the 1kGP using a method that uses linkage disequilibrium rather than quality metrics to identify 19,196 suspicious variants with 67% of them passing the 1kGP strict mask (Mafessoni et al, 2018) (Figure S15A). They identify 17,917 variants significantly associated to abnormal LD patterns that are not associated to Q.…”