A filtering strategy to reduce reference bias in measurements of allele-specific expression

Understanding how genetic variation leads to phenotypic variation is a fundamental goal in genetics. One of the challenges of this goal is to distinguish variation in a genome that has an effect on phenotype from variation that has no effect. One way to determine if a genetic variant is likely to have a phenotypic effect is to determine if that variant affects gene regulation. Presumably through their effect on gene regulation these variants affect organismal phenotypes. Examining heterozygous sites in a genome can identify the presence of one class of variants, called cisregulatory variants. In the absence of a functional variant, sequencing based measurements of gene expression should show no bias toward one allele. The presence of allelic bias implies that functional genetic variation affects gene regulation. One key issue in identifying these allele-specific events is a systematic bias toward the reference allele, an artifact of read alignment that creates false positive allele-specific expression (ASE). While N masking is one popular method used to reduce reference bias, it does not eliminate all sources of reference bias. We present a novel metric, the homology score, to characterize Single Nucleotide Polymorphisms (SNPs) based on the homology of their surrounding genomic region. When combined with N-masking, we are able to identify and filter out sources of reference bias not accounted for by use of N-masking alone. In general, this metric can assist in removing SNPs likely to yield false positive ASE.