RNA-Seq Alignment to Individualized Genomes Improves Transcript Abundance Estimates in Multiparent Populations

Including all reads by allocating allelic multireads using an EM algorithm improves the accuracy of allele-specific expression.

“Including all reads by allocating the allelic multireads using an EM algorithm improves the accuracy of allele-specific expression (Turro et al. 2011).”

• cites Turro et al. 2011 sourcePartly supported

  Materials and methods > Expectation maximization (page unavailable) describes using EM to allocate reads among transcripts, and Results > Validation of haplo-isoform deconvolution reports accurate haplotype-specific expression estimates. The paper does not establish that including all allelic multireads improves accuracy compared with excluding them; Results > MMSEQ also describes filtering some alignments and reports lower error for Gibbs-sampling estimates than for EM estimates. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

Alignment of RNA-seq reads to a reference genome can result in biased transcript abundance estimates.

“We have demonstrated that alignment of RNA-seq reads to a reference genome can result in biased transcript abundance estimates, much in the same way that a reference-based probe design can bias hybridization of genetically diverse samples to microarrays (Walter et al. 2007).”

• cites Walter et al. 2007 sourceSupported

Alignment to individualized transcriptomes improves estimation of transcript abundance.

“We demonstrate in simulated and real data sets that alignment to individualized transcriptomes increases read mapping accuracy, improves estimation of transcript abundance, and enables the direct estimation of allele-specific expression.”

• Also stated in: § Abstract · boosted

  “We demonstrate, using real and simulated data, that individualized transcriptomes improve the accuracy of read alignment and quantification.”

Variation is especially high in the three wild-derived strains (CAST, PWK, and WSB).

“Variation is especially high in the three wild-derived strains (CAST, PWK, and WSB).”

When applied to expression QTL mapping, the individualized alignment strategy corrects false-positive linkage signals.

“Moreover, when applied to expression QTL mapping we find that our individualized alignment strategy corrects false-positive linkage signals and unmasks hidden associations.”

• Also stated in: § Abstract · boosted

  “We demonstrate that in expression quantitative trait locus (eQTL) mapping, read alignment to individualized transcriptomes reduces the number of spurious linkages and unmasks extensive local genetic variation affecting gene expression.”

When applied to expression QTL mapping, the individualized alignment strategy unmasks hidden associations.

“Moreover, when applied to expression QTL mapping we find that our individualized alignment strategy corrects false-positive linkage signals and unmasks hidden associations.”

• Also stated in: § Abstract · boosted

  “We demonstrate that in expression quantitative trait locus (eQTL) mapping, read alignment to individualized transcriptomes reduces the number of spurious linkages and unmasks extensive local genetic variation affecting gene expression.”

Alignment to individualized transcriptomes enables the direct estimation of allele-specific expression.

“We demonstrate in simulated and real data sets that alignment to individualized transcriptomes increases read mapping accuracy, improves estimation of transcript abundance, and enables the direct estimation of allele-specific expression.”

The seven nonreference founder strains of the DO population differ from the reference strain C57BL/6J genomic sequence at a large number of loci.

“The seven nonreference founder strains of the DO population (A/J, 129S1/SvImJ, NOD/ShiLtJ, NZO/HlLtJ, CAST/EiJ, PWK/PhJ, and WSB/EiJ) differ from the reference strain C57BL/6J genomic sequence (NCBIM37) at a large number of loci (Table 1).”

The precision with which isoform proportions can be estimated is low with current sequencing technologies.

“However, we find that the precision with which we can estimate isoform proportions is low with current sequencing technologies and therefor focus on estimates of gene-level abundance.”

The Seqnature pipeline provides direct estimates of allele-specific expression for genes with heterozygosity.

“The Seqnature pipeline provides direct estimates of allele-specific expression for genes with heterozygosity.”

Alignment to individualized transcriptomes increases read mapping accuracy.

“We demonstrate in simulated and real data sets that alignment to individualized transcriptomes increases read mapping accuracy, improves estimation of transcript abundance, and enables the direct estimation of allele-specific expression.”

A small number of genes deviate considerably from ground truth in allele-specific expression estimates.

“A small number of genes deviate considerably from ground truth.”

Alignment of DO reads to the individualized diploid DO transcriptome improves estimates of gene abundance.

“Alignment of DO reads to the individualized diploid DO transcriptome improves estimates of gene abundance (Table 3).”

For the 439 genes, the difference in abundance estimates between alignments is less variable.

“and the difference in abundance estimates between alignments is less variable.”

There are 174 gene-level abundance estimates that differ by >50% from ground truth for the CAST alignment, vs. 485 such estimates after alignment to NCBIM37.

“There are 174 gene-level abundance estimates that differ by >50% from ground truth for the CAST alignment, vs. 485 such estimates after alignment to NCBIM37.”

A total of 5618 reads from 73 genes align to the correct unique location in the individualized DO transcriptome but align uniquely to 79 other genes in NCBIM37.

“A total of 5618 reads from 73 genes align to the correct unique location in the individualized DO transcriptome but align uniquely to 79 other genes in NCBIM37.”

Estimates of allele frequency (the proportion of reads assigned to the L allele in the DO simulation) are strongly correlated with ground truth (r = 0.82), with a median deviation of 4%.

“Estimates of allele frequency, the proportion of reads assigned to the L allele in the DO simulation, are strongly correlated with ground truth (r = 0.82), with a median deviation of 4% (Figure 2).”

Only 77 of the DO reads that improve by alignment to NCBIM37 improve by two or more categories.

“In comparison, only 5384 reads improve by alignment to NCBIM37 (Table 2, entries in boldface type) and only 77 improve by two or more categories (Table 2, entries in underlined boldface type).”

Another 137,938 reads from 5759 genes fail to align to NCBIM37 but align uniquely to the correct gene in the individualized transcriptome.

“Another 137,938 reads from 5759 genes fail to align to NCBIM37 but align uniquely to the correct gene in the individualized transcriptome (Table S4).”

Of the CAST reads that improve by alignment to NCBIM37, only 3% (326 of 10,533) improve by two or more categories.

“In comparison, only 10,533 reads improve by alignment to NCBIM37 (Table 2, entries in boldface type) of which only 3% (326 of 10,533 reads) improve by two or more categories (Table 2, entries in underlined boldface type).”

Increasing read depth does not affect these conclusions for DO.

“Increasing the read depth does not affect these conclusions (Table S5).”

Of the CAST reads that improve by alignment to CAST, 93% (427,016 of 458,297) improve by two or more categories.

“A total of 458,297 reads improve by alignment to the CAST transcriptome (Table 2, italic entries), of which 93% (427,016 of 458,297 reads) improve by two or more categories (Table 2, underlined italic entries).”

Of the 714 DO genes, 432 show more accurate estimates from individualized alignment in the simulation study.

“and most of these (432 of 714 genes) show more accurate estimates from individualized alignment in the simulation study (green circles in Figure 3C).”

Only 10,533 simulated CAST reads improve by alignment to NCBIM37.

“In comparison, only 10,533 reads improve by alignment to NCBIM37 (Table 2, entries in boldface type) of which only 3% (326 of 10,533 reads) improve by two or more categories (Table 2, entries in underlined boldface type).”

In the DO sample, 9% more reads (+704,522) align perfectly to the individualized DO transcriptome compared to NCBIM37.

“We observed that 1% more reads (+151,225) align with three or fewer mismatches and 9% more reads (+704,522) align perfectly in the individualized DO transcriptome compared to NCBIM37 (Table S6).”

These alignment errors occur between pairs or small sets of gene family members.

“These alignment errors occur between pairs or small sets of gene family members.”

Genes with retrotransposed pseudogenes are similarly sensitive to misalignment.

“Genes with retrotransposed pseudogenes are similarly sensitive to misalignment.”

Increasing the read depth does not affect these conclusions for CAST.

“Increasing the read depth does not affect these conclusions (Table S3).”

As with CAST simulated reads, polymorphic genes with closely related gene family members or retrotransposed pseudogenes are most sensitive to read alignment errors in DO.

“As with CAST simulated reads, polymorphic genes with closely related gene family members or retrotransposed pseudogenes are most sensitive to read alignment errors.”

For 2242 of the 2984 CAST genes whose abundance estimates differ >10% between alignments, simulation results show CAST alignment estimates are closer to ground truth.

“For most of this cohort (2242 of 2984 genes) the simulation results show that the CAST alignment abundance estimates are closer to the ground truth (green circles in Figure 3A).”

If only perfect-matching reads are considered, 24% of gene estimates from NCBIM37 alignment are within 10% of ground truth compared to 79% from CAST alignment.

“If we consider only perfect-matching reads (reads that align with zero mismatches), only 24% of gene estimates from the NCBIM37 alignment are within 10% of ground truth compared to 79% of gene estimates from the CAST alignment.”

A total of 458,297 simulated CAST reads improve by alignment to the CAST transcriptome.

“A total of 458,297 reads improve by alignment to the CAST transcriptome (Table 2, italic entries), of which 93% (427,016 of 458,297 reads) improve by two or more categories (Table 2, underlined italic entries).”

In the DO sample, 714 expressed genes differ by >10% in abundance estimates from alignment strategy alone.

“we find that a total of 714 expressed genes differ by >10% from alignment strategy alone”

Isoform-level alignment is also improved by alignment to CAST.

“Isoform-level alignment is also improved by alignment to CAST.”

One-quarter of expressed genes in the CAST liver sample (2984 of 11,964 genes) yield gene-level abundance estimates that differ by >10% between alignment to the CAST and NCBIM37 transcriptomes.

“One-quarter of expressed genes in the CAST liver sample (2984 of 11,964 genes) yield gene-level abundance estimates that differ by >10% between alignment to the CAST and NCBIM37 transcriptomes (Figure 3A).”

In the DO sample, 1% more reads (+151,225) align with three or fewer mismatches to the individualized DO transcriptome compared to NCBIM37.

“We observed that 1% more reads (+151,225) align with three or fewer mismatches and 9% more reads (+704,522) align perfectly in the individualized DO transcriptome compared to NCBIM37 (Table S6).”

Only 5384 simulated DO reads improve by alignment to NCBIM37.

“In comparison, only 5384 reads improve by alignment to NCBIM37 (Table 2, entries in boldface type) and only 77 improve by two or more categories (Table 2, entries in underlined boldface type).”

1119 simulated reads from Mug1 align uniquely but incorrectly in NCBIM37 to the family member Mug2.

“For example, 1119 simulated reads from Mug1 (murinoglobulin 1) align uniquely but incorrectly in NCBIM37 to the family member Mug2 (murinoglobulin 2).”

This deviation is most evident for extreme estimates of allelic imbalance.

“This is most evident for extreme estimates of allelic imbalance, which appear as horizontal lines of dots at the top and bottom of Figure 2.”

130 reads from Vcp align uniquely but incorrectly in NCBIM37 to the pseudogene Vcp-rs.

“For example, 130 reads from the protein-coding gene Vcp (valosin-containing protein) align uniquely but incorrectly in NCBIM37 to the pseudogene Vcp-rs (valosin-containing protein, related sequence).”

Most (146,083) of the DO reads that improve by alignment to the DO transcriptome improve by two or more categories.

“A total of 186,248 reads improve by alignment to the DO transcriptome (Table 2, italic entries) and most (146,083) improve by two or more categories (Table 2, underlined italic entries).”

If only perfect-matching reads are considered, 40% of gene abundance estimates from alignment to NCBIM37 are within 10% of ground truth, compared to 81% from alignment to the individualized DO transcriptome.

“If we consider only perfect-matching reads, only 40% of gene abundance estimates from alignment to NCBIM37 are within 10% of ground truth, compared to 81% of gene estimates from alignment to the individualized DO transcriptome.”

For perfectly mapping reads, 23% more CAST reads (+982,229) align perfectly to CAST than to NCBIM37.

“The difference between alignments is more striking for perfectly mapping reads (zero mismatch), as 23% more CAST reads (+982,229) align perfectly to CAST but not to NCBIM37.”

Alignment of simulated CAST reads to the CAST transcriptome improves the accuracy of gene-level abundance estimation compared to alignment to NCBIM37.

“Alignment of simulated CAST reads to the CAST transcriptome improves the accuracy of gene-level abundance estimation compared to alignment to NCBIM37 (Table 3).”

In CAST liver RNA-seq, 75% of reads align to the NCBIM37 transcriptome with three or fewer mismatches.

“We find that 75% of CAST reads align to the NCBIM37 transcriptome with three or fewer mismatches while 2.1% more reads (+252,905) align to the CAST transcriptome (Table S6).”

In CAST liver RNA-seq, 2.1% more reads (+252,905) align to the CAST transcriptome than to NCBIM37 with three or fewer mismatches.

“We find that 75% of CAST reads align to the NCBIM37 transcriptome with three or fewer mismatches while 2.1% more reads (+252,905) align to the CAST transcriptome (Table S6).”

Read alignment errors occur most frequently in genes with polymorphisms between CAST and NCBIM37, especially in genes with closely related gene family members or retrotransposed pseudogenes.

“Read alignment errors occur most frequently in genes with polymorphisms between CAST and NCBIM37, especially in genes with closely related gene family members or retrotransposed pseudogenes.”

A total of 186,248 simulated DO reads improve by alignment to the DO transcriptome.

“A total of 186,248 reads improve by alignment to the DO transcriptome (Table 2, italic entries) and most (146,083) improve by two or more categories (Table 2, underlined italic entries).”

Increasing read depth to 30 million reads and increasing stringency of filtering to require a minimum of 10 unique allelic alignments reduces the problem of extreme allele-specific expression estimates.

“Increasing read depth to 30 million reads and increasing stringency of filtering to require a minimum of 10 unique allelic alignments reduces this problem (Figure S1), arguing against widespread mappability artifacts.”

405,895 reads that failed to align to NCBIM37 are rescued to the unique, correct locus in CAST.

“We also examined 405,895 reads that failed to align to NCBIM37 but are rescued to the unique, correct locus in CAST.”

These rescued reads derive from nearly 10,000 genes (9599 genes with ≥1 reads; 4601 genes with ≥10 reads).

“These reads derive from nearly 10,000 genes (9599 genes with ≥1 reads; 4601 genes with ≥10 reads).”

Alignment to an individualized diploid transcriptome provides direct estimates of allele-specific expression.

“Alignment to an individualized diploid transcriptome provides direct estimates of allele-specific expression.”

After alignment to NCBIM37, expression of the pseudogene Rps12-ps2 (Chr 14) appears to be controlled by a distant regulator on Chr 10 at the location of the Rps12 protein coding gene.

“after alignment to NCBIM37, expression of the pseudogene Rps12-ps2 (located on Chr 14) appears to be controlled by a distant regulator on Chr 10 at the location of the Rps12 protein coding gene (Figure 4C).”

CAST- and PWK-derived reads in the individualized transcriptome align to the parent gene Rps12 rather than to the retrotransposed pseudogene Rps12-ps2.

“CAST- and PWK-derived reads in the individualized transcriptome align to the parent gene Rps12 rather than to the retrotransposed pseudogene Rps12-ps2.”

The subset of genes most sensitive to alignment method in real data consists primarily of closely related gene families and protein-coding genes with retrotransposed pseudogenes.

“The subset of genes most sensitive to alignment method in real data consists primarily of closely related gene families and protein-coding genes with retrotransposed pseudogenes.”

DO animals with the chromosome 4 region derived from the 129S1/SvImJ founder strain show higher expression of Gm12976.

“DO animals with the chromosome 4 region derived from the 129S1/SvImJ founder strain show higher expression of Gm12976 (Figure 5A, inset).”

After an individual's genetic variation is included in the alignment, particularly in the parent gene, many reads will align uniquely to the parent gene and be weighted accordingly by the EM algorithm.

“After an individual’s genetic variation is included in the alignment, particularly in the parent gene, many of these reads will align uniquely to the parent gene and be weighted accordingly by the EM algorithm (Figure S4 and Figure S5).”

Forty percent of mouse genes have multiple isoforms (15,079/37,991; median = 4 isoforms) that share substantial exon sequence.

“Forty percent of mouse genes have multiple isoforms (n = 15,079/37,991 genes; median = 4 isoforms) that share substantial exon sequence.”

Pseudogenes are particularly sensitive to false positive distant eQTL when aligned to the reference transcriptome.

“pseudogenes in particular are sensitive to false positive distant eQTL when aligned to the reference transcriptome (see explanation below).”

Summarization at the gene level with 100-bp single-end reads is robust.

“we found that summarization at the gene level with 100-bp single-end reads is robust.”

For 2900 genes, a local eQTL is revealed only after alignment to individualized transcriptomes.

“For 2900 genes, a local eQTL is revealed only after alignment to individualized transcriptomes (Figure 4A).”

Allele-specific estimates demonstrate that the local eQTL for Gm12976 is acting in cis through expression of 129S1-derived alleles in the DO animals.

“Allele-specific estimates of gene expression demonstrate that this local eQTL is acting in cis through expression of 129S1-derived alleles in the DO animals (Figure 5B).”

Alignment of simulated DO reads to individualized transcriptomes improves the accuracy of eQTL mapping relative to alignment to GRCm38.

“Alignment of simulated DO reads to individualized transcriptomes improves the accuracy of eQTL mapping relative to alignment to GRCm38.”

Among the 6097 local eQTL identified by both alignment strategies, most increase in significance from individualized alignments.

“Among the 6097 local eQTL that are identified by both alignment strategies, most increase in significance from individualized alignments (Figure 4B).”

Alignment to individualized transcriptomes improves eQTL call accuracy for all gene biotypes that exhibit eQTL differences from alignment strategy in the simulation.

“Alignment to individualized transcriptomes improves eQTL call accuracy for all gene biotypes that exhibit eQTL differences from alignment strategy in the simulation”

After alignment to individualized transcriptomes, significant local eQTL were identified for 53% of expressed genes (8997 of 16,985; FDR=1%) in real liver RNA-seq from 277 DO mice.

“We identified significant local eQTL for 53% of expressed genes (8997 of 16,985 genes expressed above threshold; FDR = 1%) after alignment to individualized transcriptomes.”

After alignment to NCBIM37, only 40% of expressed genes (6797 of 17,125; FDR=1%) had significant local eQTL.

“Only 40% of expressed genes (6797 of 17,125 genes; FDR = 1%) had significant local eQTL after alignment to NCBIM37.”

Simulations confirmed that pseudogenes can act as "read sinks" that shunt significant numbers of reads away from the protein-coding parent gene.

“Our simulations confirmed that pseudogenes can act as “read sinks” that shunt significant numbers of reads away from the protein-coding parent gene.”

In this study, eQTL mapping with gene abundance estimates derived from read alignment to the reference transcriptome missed nearly 3000 significant local eQTL associations.

“In our study, eQTL mapping with gene abundance estimates derived from read alignment to the reference transcriptome missed nearly 3000 significant local eQTL associations.”

Alignment errors to the reference transcriptome can sensitize a large group of genes to false positive local eQTL.

“our simulation study shows that alignment errors to the reference transcriptome can sensitize a large group of genes to false positive local eQTL.”

Restricting attention to only uniquely mapping reads is problematic.

“Restricting attention to only uniquely mapping reads is problematic, as we illustrate below.”

Appropriate methods for resolving the probable origin of multireads are key to obtaining accurate transcript abundance estimates.

“Appropriate methods for resolving the probable origin of multireads are key to obtaining accurate transcript abundance estimates (Mortazavi et al. 2008).”

• cites Mortazavi et al. 2008 sourceNot checked — no full text of the source

Half of all 100-bp reads from CAST should contain at least one SNP or indel.

“Thus half of all 100-bp reads from CAST should contain at least one SNP or indel.”

Expression estimates for these genes are underestimated by alignment to NCBIM37.

“Expression estimates for these genes are underestimated by alignment to NCBIM37.”

Even when the total read count is high, if the number of reads that distinguish the two alleles is low, the likelihood of sampling reads from only one of the alleles by chance can be substantial.

“Even when the total read count is high, if the number of reads that distinguish the two alleles is low, the likelihood of sampling reads from only one of the alleles by chance can be substantial.”

This extreme class of alignment error can cause major deviations in gene abundance estimates.

“This extreme class of alignment error can cause major deviations in gene abundance estimates.”

In such cases, the EM algorithm will assign all reads to the allele with the unique read alignments.

“In such cases, the EM algorithm will assign all reads to the allele with the unique read alignments.”

Simulations can identify the genes that are most sensitive to choice of alignment target.

“This suggests that simulations can identify the genes that are most sensitive to choice of alignment target;”

• AssumptionRests on an assumption: The simulation study's ground truth and results are representative enough of real RNA-seq data to identify genes sensitive to alignment target.

The simulation study appears to underestimate the actual differences in gene-level abundance estimates between alignment strategies in the DO sample.

“Again, the simulation study appears to underestimate the actual differences in gene-level abundance estimates between alignment strategies.”

• AssumptionRests on an assumption: The simulation study's ground truth and results are representative enough of real RNA-seq data to identify genes sensitive to alignment target.

The improvements in read alignment and gene-level abundance estimation are more modest in the DO sample relative to the CAST sample.

“Overall, the improvements in read alignment and gene-level abundance estimation are more modest in the DO sample relative to the CAST sample.”

• AssumptionRests on an assumption: Five of the DO founder strains are closely related laboratory strains with greater similarity to the NCBIM37 reference compared to wild-derived strains such as CAST.

In the case of Vcp-rs, the pseudogene is mistakenly classified as an expressed transcript.

“In the case of Vcp-rs, the pseudogene is mistakenly classified as an expressed transcript.”

Each DO sample will have a unique set of SNPs and indels that can bias read alignment.

“each DO sample will have a unique set of SNPs and indels that can bias read alignment”

• AssumptionRests on an assumption: Each DO sample has a unique set of SNPs and indels that can bias read alignment.

The simulation study underestimated the actual differences in gene-level abundance estimates between alignment strategies, at least for the set of 2984 genes whose abundance estimates differ by >10% between alignment strategies in the real liver data.

“this suggests that the simulation study underestimated the actual differences in gene-level abundance estimates between alignment strategies, at least for the set of 2984 genes whose abundance estimates differ by >10% between alignment strategies in the real liver data.”

• AssumptionRests on an assumption: The simulation study's ground truth and results are representative enough of real RNA-seq data to identify genes sensitive to alignment target.

Alignment to an individualized transcriptome is required to obtain accurate estimates of gene-level abundance when RNA is sequenced in genetically diverse samples.

“We have demonstrated that alignment to an individualized transcriptome is required to obtain accurate estimates of gene-level abundance when RNA is sequenced in genetically diverse samples.”

Individualized alignment improves the mapping resolution of local eQTL.

“Individualized alignment improves the mapping resolution of local eQTL.”

The problem of misalignment is amplified in genetically diverse multiparent populations like the outbred DO stock and inbred Collaborative Cross (CC) strains.

“This problem is amplified in genetically diverse multiparent populations like the outbred DO stock and inbred Collaborative Cross (CC) strains.”

Concordance between inferred founder allele effects in the DO and observed expression patterns in the founder strains provides additional evidence in support of local eQTL.

“Concordance between inferred founder allele effects in the DO and observed expression patterns in the founder strains provides additional evidence in support of local eQTL.”

For many genes, estimates of transcript abundance are relatively unaffected by alignment strategy.

“For many genes, estimates of transcript abundance are relatively unaffected by alignment strategy.”

By aligning reads from DO mice to a single search space that includes individualized sequences of both alleles for every annotated isoform, the analysis approach provides direct estimates of allele-specific expression.

“by aligning reads from DO mice to a single search space that includes individualized sequences of both alleles for every annotated isoform, our analysis approach provides direct estimates of allele-specific expression.”

In total across a large population of DO mice, the number of alignment-sensitive genes will approach the sum of all genes that are sensitive to alignment bias across all of the eight founder strains.

“in total across a large population of DO mice, the number of alignment-sensitive genes will approach the sum of all genes that are sensitive to alignment bias across all of the eight founder strains.”

Alignment to individualized transcriptomes yields new insights into gene regulation in multiparent populations.

“Alignment to individualized transcriptomes yields new insights into gene regulation in multiparent populations.”

Many of the most highly significant distant eQTL are spurious and arise from alignment errors in NCBIM37.

“we find that many of the most highly significant distant eQTL are spurious and arise from alignment errors in NCBIM37.”

Incorporation of individual genetic variation at the alignment step is critical to all downstream analyses in high throughput sequencing studies including eQTL identification.

“We have shown with simulated and real RNA-seq data that incorporation of individual genetic variation at the alignment step is critical to all downstream analyses in high throughput sequencing studies including eQTL identification.”

Gene prioritization strategies that utilize sequence data analyzed by the common reference approach will fail to identify these variants.

“Gene prioritization strategies that utilize sequence data analyzed by the common reference approach will fail to identify these variants.”

A causal variant that manifests as a local eQTL may be obscured by the common reference alignment approach.

“Thus a causal variant that manifests as a local eQTL may be obscured by the common reference alignment approach.”

Failure to incorporate individual genetic variation at the alignment step will adversely affect read alignment and consequent analyses such as abundance estimation, eQTL identification, and eQTL fine-mapping.

“In summary, failure to incorporate individual genetic variation at the alignment step will adversely affect read alignment and consequent analyses such as abundance estimation, eQTL identification, and eQTL fine-mapping.”