Scripts used to perform analysis of GWAS data in the manuscript
Bayesian clustering with uncertain data Kath Nicholls, Paul D W Kirk, Chris Wallace doi: https://doi.org/10.1101/2022.12.07.519476
The scripts were run on the university of Cambridge HPC https://www.hpc.cam.ac.uk/ using R 4.2.0 Python 3.11.0 snakemake 7.19.1
Conda environment on HPC - contains all the necessary plotting packages
conda activate basis_clustering
Download Supplementary Table 6 from Burren et al. 2020 (Genome Medicine)
wget https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7687775/bin/13073_2020_797_MOESM6_ESM.csv
Generate actual data matrices - different subsets of the data which end up in data/
Rscript prepare_data.R
Run dpMix on the datasets for various seeds, uses running_dpMix.R script and the submit_dpmix.sbatch script to submit the job using sbatch. Results end up in results/<subset>/seed<n>:
./run_all.sh
To save time reading in data, we can trim the results at this stage rather than after loading. We save these trimmed versions in trimmed_results/. In particular we discard the first half of the run and thin the samples even more, so we take only every 10th line of the file (and thus only every 1000th sample of the chain):
./trim_datasets.sh
The trimming currently isn't done for cluster parameter files, but can easily be adapted to do this too.
Main plotting scripts:
Rscript psm_plots.R
Rscript traceplots.R
Rscript quantile_traceplots.R
The quantile traceplots are faster and show a good overview of the distributions over time as long as quantiles are well separated from each other.