Impact of variant-level batch effects on identification of genetic risk factors in large sequencing studies

Abstract

Genetic studies have shifted to sequencing-based rare variants discovery after decades of success in identifying common disease variants by Genome-Wide Association Studies using Single Nucleotide Polymorphism chips. Sequencing-based studies require large sample sizes for statistical power and therefore often inadvertently introduce batch effects because samples are typically collected, processed, and sequenced at multiple centers. Conventionally, batch effects are first detected and visualized using Principal Components Analysis and then controlled by including batch covariates in the disease association models. For sequencing-based genetic studies, because all variants included in the association analyses have passed sequencing-related quality control measures, this conventional approach treats every variant as equal and ignores the substantial differences still remaining in variant qualities and characteristics such as genotype quality scores, alternative allele fractions (fraction of reads supporting alternative allele at a variant position) and sequencing depths. In the Alzheimer’s Disease Sequencing Project (ADSP) exome dataset of 9,904 cases and controls, we discovered hidden variant-level differences between sample batches of three sequencing centers and two exome capture kits. Although sequencing centers were included as a covariate in our association models, we observed differences at the variant level in genotype quality and alternative allele fraction between samples processed by different exome capture kits that significantly impacted both the confidence of variant detection and the identification of disease-associated variants. Furthermore, we found that a subset of top disease-risk variants came exclusively from samples processed by one exome capture kit that was more effective at capturing the alternative alleles compared to the other kit. Our findings highlight the importance of additional variant-level quality control for large sequencing-based genetic studies. More importantly, we demonstrate that automatically filtering out variants with batch differences may lead to false negatives if the batch discordances come largely from quality differences and if the batch-specific variants have better quality.

A Fast and Robust Strategy to Remove Variant-Level Artifacts in Alzheimer Disease Sequencing Project Data

“…18 In our exploratory analyses of these data, we observed many variants that displayed large variation in allele frequencies across centers/ platforms and contributed to spurious association signals with AD risk, that is, associations that passed at least the common suggestive significance for genome-wide association studies (p < 1 × 10 −5 ) but were of a (likely) artifactual nature. Similar to the prior study, 19 we also observed that platform/center adjustment could not fully account for these signals.…”

“…Researchers may thus inspect filtered variants in targeted analyses in subsets of the ADSP data where no artifactual genotype enrichment is observed (e.g., excluding a single sequencing center/ platform that showed an artifactual increase in genotype counts compared with the others, cf. Wickland et al 19 ).…”

“…However, a prior study using a prior version of the ADSP WES discovery phase observed that center/platform covariate adjustment could not account for variable variant qualities across centers and platforms, which in turn may lead to spurious associations or affect the identification of AD-associated risk variants. 19…”

“…However, a prior study using a prior version of the ADSP WES discovery phase observed that center/platform covariate adjustment could not account for variable variant qualities across centers and platforms, which in turn may lead to spurious associations or affect the identification of AD-associated risk variants. 19 Since then, the ADSP has further expanded its efforts and as of 2021, provides their WES and WGS data sets on 20.5 and 16.9 k individuals, respectively, across diverse ancestries. 18 In our exploratory analyses of these data, we observed many variants that displayed large variation in allele frequencies across centers/ platforms and contributed to spurious association signals with AD risk, that is, associations that passed at least the common suggestive significance for genome-wide association studies (p < 1 × 10 −5 ) but were of a (likely) artifactual nature.…”

Genome-wide analyses of early-onset acute myocardial infarction identify 29 novel loci by whole genome sequencing

“…In the present study, WGS data were analyzed to characterize the entire genome to identify early-onset AMI-specific markers, revealing 86% loci and 77% variants associated with early AMI that were not covered in SNP chips (Supplementary Material, Table S2 and S3). Moreover, the variants were strictly filtered to remove false positives caused by the batch effect which can lead to false disease association (36, 37). A batch effect was confirmed according to the sequencing year, and it was not successfully filtered by common variant filtering criteria such as missing genotype rate, minor allele frequency, and Hardy-Weinberg equilibrium (Supplementary Material, Fig.…”