Repository for the analyses done in the publication "Aberrant splicing prediction across human tissues".

The project is setup as a snakemake pipeline.

• The workflow folder contains the config files and the analysis scripts for each dataset.
  • config contains yaml files for the GTEx, mito and ALS datasets, as well as a yaml that defines the file structure used in all analysis.
  • scripts contains all processing and analysis scripts. The common folder contains all scripts shared across datasets. gtex_v8, mito and als contain analysis scripts specific to the respective dataset. Each of the dataset folders contains a Snakefile that runs the analysis of the respective dataset. As mito and als depend on gtex_v8 results (e.g. AbSplice model, SpliceMaps from GTEx tissues), the gtex_v8 Snakefile needs to be executed first.

In scripts/common most of the scripts of the pipeline are included. The main tasks of the pipeline are:

• vcf_annotation filters the provided vcf file for rare, high quality variants
• junction_annotation generates SpliceMaps based on RNA-seq split read counts from FRASER
• outlier_ground_truth generates FRASER outliers on gene and junction level, filtered for significance and rare variants in the vicinity
• splicing_pred generates splicing predictions for SpliceAI, MMSplice, MTSplice, CADD-Splice, SQUIRLS and AbSplice
• splicing_result provides maximum predictions on a gene and variant level
• benchmark creates the benchmark dataframes from predictions and outliers. Where each gene, sample, tissue combination in the benchmark contains at least one rare variant on the gene

In order for the pipeline to work users need to provide some files:

• outlier calls from DROP. Create a symlink to the DROP results in the folder ./workflow/data/resources/{dataset}/DROP/, where dataset can be gtex_v8, mito and als.
  We used DROP v.1.1.2. For GTEx v8 the necessary config file to run DROP is provided in here: ./workflow/data/resources/gtex_v8/DROP_config.yaml, with the corresponding sample annotation for DROP in here: ./workflow/data/resources/gtex_v8/DROP_sample_annotation.tsv.

• rocksdb database with precomputed SpliceAI scores. Create a symlink in ./workflow/data/resources/common/{genome}/SpliceAI/spliceai.db, where genome can be hg19 and hg38. These databases can be created from: https://github.com/gagneurlab/spliceai_rocksdb.

• rocksdb database with gnomAD minor allele frequencies. Create a symlink in ./workflow/data/resources/common/{genome}/gnomAD_maf_db/rocksdb/maf.db, where genome can be hg19 and hg38. These databases can be created from: https://github.com/gagneurlab/gnomad_rocksdb.

• vcf files from the dataset. Users need to normalize the vcf files and store them in ./workflow/data/resources/{dataset}/vcf_normalized/{vcf_id}.vcf.gz, where dataset can be gtex_v8, mito and als. vcf_id can be anything (e.g. sampleID or split huge vcf by chromosome to parallelize computations).

Follow these steps to run the pipeline:

• Install the absplice environment. Follow the steps indicated in the absplice GitHub. You can use the docker image or conda environment.
• Activate the absplice environment.
• From the absplice environment, run the snakemake workflow for gtex_v8 (below just with 1 job; increase this depending on your available resources):

  cd workflow/scripts/gtex_v8
  python -m snakemake -j 1 --use-conda
  

• Similarly run the snakemake workflow for ALS.
• The mito dataset is not publicly available.

All figures from the manuscript can be generated from the the provided scripts in here. The minimal dataset to produce the figures is available here. To run the scripts, first download and unzip the provided data folder to the root directory of this repository.