flexiMAP: a regression-based method for discovering differential alternative polyadenylation events in standard RNA-seq data

Abstract

Motivation We present flexiMAP (flexible Modeling of Alternative PolyAdenylation), a new beta-regression-based method implemented in R, for discovering differential alternative polyadenylation events in standard RNA-seq data. Results We show, using both simulated and real data, that flexiMAP exhibits a good balance between specificity and sensitivity and compares favourably to existing methods, especially at low fold changes. In addition, the tests on simulated data reveal some hitherto unrecognised caveats of existing methods. Importantly, flexiMAP allows modeling of multiple known covariates that often confound the results of RNA-seq data analysis. Availability The flexiMAP R package is available at: https://github.com/kszkop/flexiMAP Scripts and data to reproduce the analysis in this paper are available at: https://doi.org/10.5281/zenodo.3689788 Supplementary information Supplementary data are available at Bioinformatics online.

Polygenic risk for schizophrenia converges on alternative polyadenylation as molecular mechanism underlying synaptic impairment

“…From these values, r prox and r dist , we computed the percentage of distal 3’UTR used for each gene j and each iN donor I , analogous to previous studies 66,67 as R ij = r distij /(r proxij +r distij ). To quantify differential PAS site usage between HC and SCZ in iNs, we employed a beta regression approach as reported previously using a logit link function for the mean model, the identity function for the precision model as well as a maximum likelihood estimator 66 , correcting for gender, batch and TechnicalPC1 (see the RNA-Seq analysis), testing for diagnosis status. Similarly, we also corrected the R ij values for gender, batch and TechnicalPC1 using beta regression and used the residuals to compute the average difference in R values between the groups (ΛAPA, Supplementary Table 3 ).…”

“…For differential APA analysis, we next quantified the normalized read count in all full-length RNA-Seq samples for each of the non-overlapping 3’UTR regions for each gene, counting all reads overlapping with the proximal or distal UTR region respectively. From these values, r prox and r dist , we computed the percentage of distal 3’UTR used for each gene j and each iN donor I , analogous to previous studies 66,67 as R ij = r distij /(r proxij +r distij ). To quantify differential PAS site usage between HC and SCZ in iNs, we employed a beta regression approach as reported previously using a logit link function for the mean model, the identity function for the precision model as well as a maximum likelihood estimator 66 , correcting for gender, batch and TechnicalPC1 (see the RNA-Seq analysis), testing for diagnosis status.…”