Effect of mapping bias on eQTL discovery. (A) Comparison of original and filtered P values (rho =0.92, P value <2.2e-16) shows that for the vast majority of the genes, the P values after filtering potentially biased reads are highly consistent with P values without filtering. Colors denote whether the gene with an eQTL in the original dataset was significant only before (lost) or only after (gained) filtering putatively biased reads or in both analyses (common). The dotted lines denote 10% FDR significance thresholds. (B) The number of exons per gene with significant associations as a function of the total number of quantified exons in the original, non-filtered dataset. (C) Proportion of the best-associated exon per gene with genetic variants (see also Additional file 1: Table S5). (D) Proportion of biased variants in six different categories based on the single-end genome based mapped with BWA simulated reads. Matched eQTL null is a random sample of variants matched to the distance from TSS of eQTLs.