Lack of Details About Input and Output Data

[DarioS]

• Should the variants file be subset to only the regions of interest for speed or will the script do it at the beginning of execution?
• What is the recommended minimum number of biological samples per run? Can it run in single-sample mode?
• Example file SEISMIC_demo_UCEC_WXS_result__unadjusted_bkg_mutrate_cohort_sig_UCEC_min_3_muts_10000_sims.tsv seems to be in terms of regions. Is SEISMIC able to predict specific driver SNPs for experimental investigation by CRISPR cell line editing?

It would be nice to see these points addressed in the usee guide for the benefit of beginner users.

[m-bostrom]

• Mutations are subset right at the beginning of the script, yes, so I would not bother doing it manually. The analysis is slow enough that a small difference in initial mutation loading won't make a big difference in speed. Also, subsetting to a smaller region than what the mutational models are calculated from (The whole genome for WGS, regions in wxs_covered_regions_path for WXS) would impact model performance.
• There is currently no recommended minimum number of samples. It would depend on other factors as well, such as the spread of mutation burden among them. But I've generally seen better results with hundreds than with tens of samples. A single-sample mode would not work, as the whole concept is to check whether a cohort's mutations in a region occur in unlikely samples. With a single sample, the probability that any observed mutations are in that sample would be 100%.
• The analysis is entirely region-based. For base-level rather than region-level analysis, I would generally recommend using other tools. However, if you are interested in a particular position that is known ahead of time, you could define a region that only covers that position. The path to the file with regions to be tested is set in test_regions_path - the supplied file can be used for reference regarding the format. Testing individual positions like this has not been tested extensively though, so while it works in theory I can't speak to performance. Alternatively, the plots that are generated for regions under a certain p value (plotting_max_p) or that are requested manually (plot_region_list) do show the distribution of mutations in the region, but there are no statistical tests regarding their positions.

I will try to find the time to update the README to address some of these points.