Skip to content

Latest commit

 

History

History
208 lines (140 loc) · 9.02 KB

README.md

File metadata and controls

208 lines (140 loc) · 9.02 KB
Raw

Code and data for "Conditional prediction of consecutive tumor evolution using cancer progression models: What genotype comes next?"

Juan Diaz-Colunga, Ramon Diaz-Uriarte

Published in PLoS Computational Biology, https://doi.org/10.1371/journal.pcbi.1009055

Main scripts and directories

Here we provide an overview of the main scripts and directories. Other files and directories will be mentioned, as needed, below. Where deemed of used for readers, we also provide .Rout files. RData and rds files are provided when their size permits.

Executing the following scripts in the indicated order and with default options creates a ./data directory within the current one where transition matrices extracted from both the OncoSimulR simulations and the CPMs are stored. It also generates a table.RData object containing comparisons across said matrices.

oncoFunctions.R - Contains functions called by many other scripts and some tests.

next-genotype_transitionMatrix-sim.R - Extracts transition matrices directly from simulations. By default, matrices are stored in the ./sim directory.

next-genotype_transitionMatrix-cpm.R - Extracts transition matrices from the outputs of the CPMs. By default, matrices are stored in the ./cpm directory.

next-genotype_transitionMatrix-timeDiscretizedCPMs.R - Works just like the previos script, but for Schill's MHN and the time-discretized versions of CBN_ot and MCCBN. Output matrices are also stored under the ./cpm directory by default.

structData.R - Takes the matrices under the ./sim and ./cpm directories (by default, but different directories can be provided) and groups them into nested lists stored in .RData objects and saved in the ./data directory (by default).

makeTable.R - Reads the content of the ./data directory (by default) and creates the table.rds file containing a data frame with the comparative statistics across transition matrices.

oncoPlots.R - Sourcing this file provides a set of functions that can be used to produce plots taking the table generated by makeTable as input. Not used.

merge-additional-info.R - Run after makeTable.R. Merges additional info and outputs table-replicates.rds (approx. 20 GB file).

average-and-array-statistics.R - Run after merge-additional-info.R. Computes per-array statistics (with different weighting schemes) and averages over replicates, genotypes, and number of mutations, appropriately.

makeTableObservedFreqs.R - Can be run at any time, but necessarily before merge-additional-info.R. Returns observed frequencies of genotypes in actual samples used.

makeTableRanksFitness.R - Can be run at any time, but necessarily before merge-additional-info.R. Returns fitness ranks of genotypes in fitness landscapes.

makeTableLocalMax.R - Can be run at any time, but necessarily before merge-additional-info.R. Returns frequency with which each genotype is a local maximum in simulations.

makeTableWeights.R - Can be run at any time, but necessarily before merge-additional-info.R. Returns the true frequency with which a genotype is the most abundant genotype during the simulations, as well as the frequency of each genotype in the complete 20000 samples. These two are the two sets of weights used.

makeTable-to-diff-wrt-null-2.R - Computes difference in JS between the null model and the method output.

paired-wide-data.R - Created wide (or paired-wide) data frames, so finding out best method or performing paired tests is easier.

create_rank_data.R - Rank methods and obtain the best method in each case. Requires wide data. Comparisons made at the number mutations, array, and single genotype (per replicate) case.

create_wide_g_long_g.R - Create minimal, concise and with factors, data sets for the glmertree models.

data_for_weighted_glmertree_plots.R - (under plots-glms) Create data set to produce plots from lmertree fits. This produces ""data_for_weighted_glmertree_plots.RData".

merge-compsfixNulls-wide_g.R and explore-reduce-method-comp.R: comparisons among methods: merging with wide_g data for lmertree fits. For some tables below. (E.g., in file when_good_conditions_0677.R)

method-comp-biol-data.R Comparison of predictions between methods on the biological data, under /run-biol-examples.

./plot-glms Directory with code for fitting the lmertrees and plotting the results. Additionally, also scripts for producing tables in Supplementary material of performance when similar CE and TD (when_good_conditions_0677.R)

./run-biol-examples Code and data to run the CPM analyses and obtain the transition matrices on the cancer data sets.

./biol-data-plots Plots in manuscript and supplementary material for the output from the analyses on the cancer data sets.

./Schill-MHN Code from Schill et al., for MHN plus additional code for us to obtain transition matrices. Since there is no license in the original repository (https://github.com/RudiSchill/MHN) we have not provided it. The code should be left under Shcill-MHN/MHN (without creating a new subdirectory). A README.txt further explains the files and functions under this directory.

CBN-with-bug-fix-and-likelihood A directory that contains the compressed directory of H-CBN with a couple of bug fixes and enhancements by RDU, as explained in the Supplementary Material. The original code from https://bsse.ethz.ch/cbg/software/ct-cbn.html is under the GNU GPL (v. 3), the same used here. The authors of the code are Niko Beerenwinkel, Moritz Gerstung, and Seth Sullivant.

Fitnesslandscape-characteristics Code for plots of fitness landscapes characteristics. One of them is used in the Supplementary Material (gamma-rsign-obs-peaks-FL.pdf)

Pipeline

  1. The time discretized CBN and MCCBN transition matrices as well as MHNs are needed for analyses below. (Before the next-genotype-... are run). Run Time-discretized-CBN-MCCBN/time-discretized-CBN.R and Schill-MHN/run-Schill-MHN-trans-mat.R

  2. At any time, but before running merge-additional-info.R, run each of makeTableObservedFreqs.R, makeTableRanksFitness.R, makeTableLocalMax.R, and makeTableWeights.R.

    Output:

    • makeTableObservedFreqs.R: outObsFreqs.RData (~ 95 MB, but fast script)

    • makeTableRanksFitness.R: outRanksFitness.RData. (~ 43 MB, but fast script)

    • makeTableWeights.R: weightsAll.RData [plus intermediate RData: outfrerqsSampled and outFrerqsTrue]. (~ 17 MB; takes about 6 h)

    • makeTableLocalMax.R: outLocalMax.RData (~ 560 KB; takes about 7 h)

  3. Run all three scripts next-genotype_transitionMatrix-*

  4. Run structData

  5. Run makeTable.R. The -scratch flag is needed the first time this script runs, then once the temporary file makeTable_tmp.rds file has been created, formatting changes can be made and implemented by running makeTable with no flags. If any upstream changes are made, a new -scratch run will be needed to incorporate them.

  6. Run merge-additional-file-info.R. Input: pre-table-replicates.RData (plus fl_sampl.RData and the output from each of the files run in step 0). Output: table-replicates.rds.

  7. Run average-and-array-statistics.R. Input: table-replicates.rds. Output: FIXME: zz, spell out.

  8. At this point, the generated tables can be used for downstream analyses and plots.

  9. Run makeTable-to-diff-wrt-null-2.R: Input: data_no_any_over_replicate.rds, array_statistics_over_replicate.rds, data_no_any.rds, num_mut_statistics_over_replicate.rds. Output: genotype_diff_null.RData, array_diff_null.RData, genotype_diff_null_no_average.RData, num_mut_diff_null.RData.

  10. Run paired-data.R. Input: genotype_diff_null.RData, array_diff_null.RData, genotype_diff_null_no_average.RData, num_mut_diff_null.RData. Output: wide_array.RData, wide_genotype.RData, wide_genotype_no_average.RData, wide_num_mut.RData.

  11. Run create_rank_data.R, then create wide_g_long_g.R. Note that we eliminate cases with sampledProp == 0 (create_wide_g_long_g.R), then those with sampledProp < 1e-3 (fit-lmtree scripts and modelse-beta-regression.R), as well as the last genotype ---7 mutations or 10 mutations, for landscapes with 7 and 10 loci, respectively--- (fit-lmtree scripts and modelse-beta-regression.R).

  12. Run the lmertree fits. In subdirectory plots-glms/fits-Weighted-01-sqrt. A launch.sh will launch the lmertree fits. Once done, plotting with plotting-lmertree-fits-weighted-01-sqrt.R

  13. Analyze the cancer data sets. Under ./run-biol-examples

    • Launch analyses with CPMs. See details under README.txt. File launch-run-rerun.sh launches all R processes.

    • Once analyses with CPMs are completed, compare output: method-comp-biol-data.R

  14. Run the two files under ./biol-data-plots to generate the plots for the cancer data sets (main manuscript and supplementary material).

Contact

For questions or comments, please contact J. Diaz-Colunga or R. Diaz-Uriarte.