This repository will recreate the poster used at the first GECIP conference and provide the associated information required for reproducibility.
The poster was created using the posterdown package. To run, render the Posterdown.Rmd file. The output will be in html format. A PDF version is also provided in this repo. If you don't have the required libraries please run Rscript install.R from the repository directory.
The SEPATH pipeline is available as a singularity on github.
This pipeline was ran on all cancer whole genome sequences (v6 release) from the 100,000 Genomes Project with the following parameters (all others parameters were as default):
krakendb - All genomes in NCBI Refseq scaffold level and above (bacterial, viral, fungal, protozoal). Dustmasked to remove low complexity. See resources/names.dmp and resources/nodes.dmp for taxonomy information. database.kdb MD5:1e5bbafdef19775b6a9a9055598fb709
kraken_confidence - 0.2 - 20% of assigned k-mers within a read must assign to a taxonomy before classification can be determined
min_clade_reads - 0 - left unfiltered until analysis
bbduk_db - Human reference genome 38 (no decoys) with additional cancer sequences from the COSMIC database
minimum_quality - 20 - in addition to quality control from Illumina
minimum_length - 35 - sufficient for k=31 runs with kraken. set low to preserve data
Sample metadata was accessed via the R labkey API in Genomics England Research Environment. Main programme data used was for v7 data release as of 2019-07-25. Associated scripts for data cleaning and formatting can be provided to GeCIP members from within the research environment upon request.
Community matrix filtering:
- All samples were aligned to genome build GRCh38, passed Illumina internal QC, were obtained from fresh-frozen tissue and library prep was PCR-free
- Taxa within samples with less than 50 assigned sequencing reads were set to 0 to reduce false positives
- Converted to binary with
kraken_pa <- decostand(kraken_prep, method='pa')
PCoA:
- Distance matrix obtained with
jacc <- vegdist(kraken_pa, method='jaccard', binary=TRUE) - Multidimensional scaling performed with
cmdscale(jacc, k=2, eig=TRUE, x.ret=TRUE) - vegan pacakge version:
2.5.3 - Data points and principal MDS axes variance were extracted, remerged with metadata by sequencing plate_key and plotted with ggplot.
Scripts available from within research environment. Boruta settings required for reproducibility:
- R version:
3.5.1 - Random seed setting:
set.seed(1122) - Boruta package:
6.0.0 - Boruta Parameters:
maxRuns=500. All other parameters as default
The importance decisions for each boruta run were extracted for each technical variable. The combined number of times a genus was confirmed as important for predicting a technical variable was ranked the the top 15 decisions were compared to table 1 in Eisenhofer et al 2019. The resulting dataframe is provided in code of the poster markdown file.