pipeline_slamdunk_umis

The pipeline performs the following:

• deduplicate reads with umi_tools
• Performs SLAM-seq analysis with SLAM-DUNK Docs: https://t-neumann.github.io/slamdunk/docs.html#docstart
• Determine mRNA half-life based on Tcounts output from SLAM-DUNK

Inputs needed

1. Inputs Reads Input reads must be the lone present in the cwd with suffix ".fastq.gz" Input reads names should have the following structure: trimmed-cellline-timepointinhours-R#(replicate number) example: trimmed-cho-3h-R1.fastq.gz
2. A genome file
3. A bed file of 3'UTRs

The pipeline outputs the different files/directories

• sample_description.tsv file with sample names and time points asked by slamdunk_all
• xxx_processed.fastq files : ouputs of the umi tools extract
• map directory : contains outputs from slamdunk map "xxx_slamdunk_mapped.bam" and .log
• filter directory : contains output from
  • slamdunk filter "xxx_slamdunk_mapped_filtered.bam" with index .bai and .log
  • umi_tools dedup "xxx_slamdunk_mapped_filtered_dedup.bam" with index .bai
  • slamdunk alleyoop summary "xxx_slamdunk_mapped_filtered_summary.tsv" and .log, "xxx_slamdunk_mapped_filtered_summary_PCA.txt"
• snp directory : contains outputs from slamdunk snp "xxx_slamdunk_mapped_filtered_snp.vcf" and .log
• count directory : contains outputs from
  • slamdunk count "xxx_slamdunk_mapped_filtered_dedup_tcount.tsv" and .log with Conversion rates for each transcript (used to determine half-life), "xxx_slamdunk_mapped_filtered_dedup_tcount_mins.bedgraph" and "xxx_slamdunk_mapped_filtered_dedup_tcount_plus.bedgraph"
  • cgat combine_tables "xxx-aggConvRate.tsv" and "xxx-agg-ReadsCPM.tsv"
• halflife directories contains output of Rscripts
  • ConvRate_processed.tsv : Conversion Rates table of each sample and replicate, normalized to time-point 0h and background substracted when no4su samples are provided
  • halflife_unfiltered.csv : selected halflife between the 3 models half-life were not filtered meaning abberant half-life are present OR here only model 1 hal-lifes
  • model_fitting_summary.txt : summary of half-life below 0h or above the max time-point OR NOT if only model 1
  • models_halflife_decay_aic.tsv: result of all models if 3 models, nothing if only model 1
  • bootstrap_halflife.tsv : bootstrapped half-lifes (iteration = 1000)
  • halflife_percentileCIs.tsv : Pencentile CI calculated on bootstrapped half-lifes, wth other stats like mean and median
  • halflife_filtered.tsv : filtered half-lifes for h-l that were out of the bs CI, < 0h or > 24h
  • halflife_filtered.log : number of transcripts filtered

Requirements

On top of the default CGAT setup, the pipeline requires the following

• Software:
  • python (v3.8.12 with pysam v0.17.0 when built)
  • slamdunk (v0.4.3 when built)
  • samtools (v1.12 when built)
  • umi_tools (v1.0.1 when built)
  • meme (v5.3.0 when built)
• R modules:
  • optparse
  • matrixStats
  • tidyverse
  • stringr
  • foreach
  • doParallel
• Python modules
  • pyteiser (pip install) https://github.com/goodarzilab/pyteiser

Configuration

The pipeline requires a configured :file: pipeline.yml file.

Make a directory with your project name. Configure the pipeline with python [path_to_repo]/pipeline_slamdunk_umis.py config. A pipeline.log and pipeline.yml file(s) will be added to your new directory. Modify the pipeline.yml according to your project (specify annotation database and directory, database for uploading the outputs; specify options for Salmon quantification).

Pipeline use

Run the pipeline with python [path_to_repo]/pipeline_slamdunk_umis.py make full -v5.

For running the pipeline on a large set of samples, submit the pipeline onto the cluster (sharc), using a submit_pipeline custom script.

P.S

/!/!\Only model 1 included in pipeline/!/!
The Rscript directory contains 4 scripts to calculate h-l from conversion rates. all rep aboceCPM: model(s) are given T>C Conversion rates of transcripts with a CPM above threshold in all replicates of each time point. 2rep aboveCPM: model(s) were given model(s) are given T>C Conversion rates of transcripts with a CPM above threshold in at least 2 replicates of each time points

3model: Rscript with the 3 models: these scripts are not included in the pipeline, but I kept the code in case.

model1: Rscript with only model 1.