Workflows used for Teran et. al. "Nonsense-mediated decay is highly stable across individuals and tissues"

These scripts constitute the main workflows used for Teran et. al. "Nonsense-mediated decay is highly stable across individuals and tissues": https://doi.org/10.1101/2021.02.03.429654

Variant annotation - annotate_variants

Variant annotation was performed the Variant Effect Predictor (https://genomebiology.biomedcentral.com/articles/10.1186/s13059-016-0974-4, https://github.com/Ensembl/ensembl-vep), including the LOFTEE plugin (https://github.com/konradjk/loftee). Bash scripts used to run VEP are provided, along with an R script used to aggregate the annotations. Ensembl 88 was used to obtain all annotations except GnomAD allele frequencies, which where obtained from Ensembl 97.

Predictive modeling - predictive_models

NMD was predicted using penalized logistic regression. Each subfolder contains a model with different predictors: lindeboom - only the predictors from NMDetective (https://www.nature.com/articles/s41588-019-0517-5). full - the predictors from lindeboom in addition to conservation scores, GC content, RNA metrics and allele frequencies. conditional - the same as full but only in variants predicted to undergo NMD based on the 50bp rule. subject - the predictors in full along with GTEX subject ID. tissue - the predictors in full along with tissue name.

The comparison subfolder contains the code used to compare the difference in performance across the models.

Multitissue ASE modeling and cross-tissue ASE correlations - multitissue_ase

Multitissue ASE probabilities were modeled using the R code in ASE_29Dec2014_new.R. The other R scripts in the multitissue_ase folder were used to call this code to assign ASE probabilities to all ASE measurements.

The ase_correlation subfolder contains the code to compute pairwise correlations of ASE across all pairs of tissues. All variants in all subjects and tissues quantified for ASE with GATK ASEReadCounter and adjusted with WASP (https://github.com/bmvdgeijn/WASP) were reannotated used VEP with Ensembl 88. Variants in the categories stop_gained, missense_variant, synonymous_variant, intron_variant, 3_prime_UTR_variant, 5_prime_UTR_variant, and non_coding_transcript_exon_variant and files were aggregated by tissue instead of subject. (Note: this part is very computationally intensive and should probably be run on a server).