Abstract
Allele-specific expression (ASE) captures the functional impact of genetic variation on transcription, offering a high-resolution view of cis-regulatory effects, but its quality can be diminished by technical, biological, and analysis artifacts. We introduce aseQC, a statistical framework that quantifies sample-level ASE quality in terms of the overall expected extra-binomial variation to exclude uncharacteristically noisy samples in a cohort to improve robustness of downstream analyses. Applying aseQC to a dataset of rare mendelian muscular disorders, successfully identified previously annotated low-quality cases demonstrating clinical genomic utility. When applied to 15,253 samples in extensively quality controlled GTEx project data, aseQC uncovered 563 low-quality samples that exhibit excessive allelic imbalance. We identify these to be associated with specific processing dates but not otherwise described adequately by any other quality control measures and metadata available in GTEx data. We show that these low-quality samples lead to 23.6 and 31.6 -fold increased ASE, and splicing outliers, degrading the performance of transcriptome analysis for rare variant interpretation. In contrast, we did not observe any adverse effect associated with inclusion of these samples in common-variant analysis using quantitative traits loci mapping. By enabling quick and reliable assessment of sample quality, aseQC presents a critical step for identifying subtle quality issues that remain critical for a successful analysis of rare variant effects using transcriptome data.