Limits...
MetaDiff: differential isoform expression analysis using random-effects meta-regression.

Jia C, Guan W, Yang A, Xiao R, Tang WH, Moravec CS, Margulies KB, Cappola TP, Li M, Li C - BMC Bioinformatics (2015)

Bottom Line: Detection of differential isoform expression is challenging because of uncertainty in isoform expression estimation owing to ambiguous reads and variability in precision of the estimates across samples.It is desirable to have a method that can account for these issues and is flexible enough to allow adjustment for covariates.In contrast, several existing methods either fail to control false discovery rate or have reduced power in the presence of covariates or confounding variables.

View Article: PubMed Central - PubMed

Affiliation: Department of Biostatistics and Epidemiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, 19104, USA. jiacheng@mail.med.upenn.edu.

ABSTRACT

Background: RNA sequencing (RNA-Seq) allows an unbiased survey of the entire transcriptome in a high-throughput manner. A major application of RNA-Seq is to detect differential isoform expression across experimental conditions, which is of great biological interest due to its direct relevance to protein function and disease pathogenesis. Detection of differential isoform expression is challenging because of uncertainty in isoform expression estimation owing to ambiguous reads and variability in precision of the estimates across samples. It is desirable to have a method that can account for these issues and is flexible enough to allow adjustment for covariates.

Results: In this paper, we present MetaDiff, a random-effects meta-regression model that naturally fits for the above purposes. Through extensive simulations and analysis of an RNA-Seq dataset on human heart failure, we show that the random-effects meta-regression approach is computationally fast, reliable, and can improve the power of differential expression analysis while controlling for false positives due to the effect of covariates or confounding variables. In contrast, several existing methods either fail to control false discovery rate or have reduced power in the presence of covariates or confounding variables. The source code, compiled JAR package and documentation of MetaDiff are freely available at https://github.com/jiach/MetaDiff.

Conclusion: Our results indicate that random-effects meta-regression offers a flexible framework for differential expression analysis of isoforms, particularly when gene expression is influenced by other variables.

No MeSH data available.


Related in: MedlinePlus

Quantile-quantile (QQ) plots of different tests in detecting DE isoforms. Displayed are p-values for those true non-DE isoforms
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4489045&req=5

Fig4: Quantile-quantile (QQ) plots of different tests in detecting DE isoforms. Displayed are p-values for those true non-DE isoforms

Mentions: We also examined the QQ plots for non-DE transcripts (Fig. 4). A good-performing method is expected to have –log10 transformed p-values falling along the diagonal line in a QQ plot. In general, BcLR and t-test had p-values close to the expected distribution in all three scenarios. However, some of the other methods had strong deviation from the diagonal line even though their empirical FDRs were under control. EdegR showed upward deviation in Scenario I when sample size was small (m = 4) or moderate (m = 8). DESeq and DESeq2 tended to deviate upward in all three scenarios, and the degree of deviation was more pronounced in Scenario I. Such deviation for EdgeR, DESeq, and DESeq2 is likely due to their inability to account for isoform expression estimation uncertainty. In Scenario III, Cuffdiff showed a substantial upward deviation from the diagonal line, which is consistent with its highly inflated FDRs shown in Fig. 3. The plateau is due to the sampling based method employed by Cuffdiff for significance evaluation; the current program only gives p-values ≥ 5 × 10−5.Fig. 4


MetaDiff: differential isoform expression analysis using random-effects meta-regression.

Jia C, Guan W, Yang A, Xiao R, Tang WH, Moravec CS, Margulies KB, Cappola TP, Li M, Li C - BMC Bioinformatics (2015)

Quantile-quantile (QQ) plots of different tests in detecting DE isoforms. Displayed are p-values for those true non-DE isoforms
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4489045&req=5

Fig4: Quantile-quantile (QQ) plots of different tests in detecting DE isoforms. Displayed are p-values for those true non-DE isoforms
Mentions: We also examined the QQ plots for non-DE transcripts (Fig. 4). A good-performing method is expected to have –log10 transformed p-values falling along the diagonal line in a QQ plot. In general, BcLR and t-test had p-values close to the expected distribution in all three scenarios. However, some of the other methods had strong deviation from the diagonal line even though their empirical FDRs were under control. EdegR showed upward deviation in Scenario I when sample size was small (m = 4) or moderate (m = 8). DESeq and DESeq2 tended to deviate upward in all three scenarios, and the degree of deviation was more pronounced in Scenario I. Such deviation for EdgeR, DESeq, and DESeq2 is likely due to their inability to account for isoform expression estimation uncertainty. In Scenario III, Cuffdiff showed a substantial upward deviation from the diagonal line, which is consistent with its highly inflated FDRs shown in Fig. 3. The plateau is due to the sampling based method employed by Cuffdiff for significance evaluation; the current program only gives p-values ≥ 5 × 10−5.Fig. 4

Bottom Line: Detection of differential isoform expression is challenging because of uncertainty in isoform expression estimation owing to ambiguous reads and variability in precision of the estimates across samples.It is desirable to have a method that can account for these issues and is flexible enough to allow adjustment for covariates.In contrast, several existing methods either fail to control false discovery rate or have reduced power in the presence of covariates or confounding variables.

View Article: PubMed Central - PubMed

Affiliation: Department of Biostatistics and Epidemiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, 19104, USA. jiacheng@mail.med.upenn.edu.

ABSTRACT

Background: RNA sequencing (RNA-Seq) allows an unbiased survey of the entire transcriptome in a high-throughput manner. A major application of RNA-Seq is to detect differential isoform expression across experimental conditions, which is of great biological interest due to its direct relevance to protein function and disease pathogenesis. Detection of differential isoform expression is challenging because of uncertainty in isoform expression estimation owing to ambiguous reads and variability in precision of the estimates across samples. It is desirable to have a method that can account for these issues and is flexible enough to allow adjustment for covariates.

Results: In this paper, we present MetaDiff, a random-effects meta-regression model that naturally fits for the above purposes. Through extensive simulations and analysis of an RNA-Seq dataset on human heart failure, we show that the random-effects meta-regression approach is computationally fast, reliable, and can improve the power of differential expression analysis while controlling for false positives due to the effect of covariates or confounding variables. In contrast, several existing methods either fail to control false discovery rate or have reduced power in the presence of covariates or confounding variables. The source code, compiled JAR package and documentation of MetaDiff are freely available at https://github.com/jiach/MetaDiff.

Conclusion: Our results indicate that random-effects meta-regression offers a flexible framework for differential expression analysis of isoforms, particularly when gene expression is influenced by other variables.

No MeSH data available.


Related in: MedlinePlus