/PU-learning-paper-analysis

Data and analysis scripts for the PU learning paper

Primary LanguagePostScript

Inferring protein sequence-function relationships with large-scale positive-unlabeled learning

Folder Structures

data-r/

Takes input protein files from data/ folder, and generate rdata files for each protein.

  • Input: data/[protein]/[protein]_ref_sequences_filtered.txt.gz, data/[protein]/[protein]_sel_sequences_filtered.txt.gz
  • Output: [protein].rda

code/

Related to fitting models,

code/vfits

  • R scripts: vfits.R, vfits_i_1.R, vfits_i_2.R

    • vfits.R contains a script fitting PU models over 20 py1 values, where at each py1 value 10-fold CV are performed.
    • vfits_i_1.R is a parallel version of vfits.R. It contains a script fitting PU models for one CV fold.
    • vifts_i_2.R is a script which aggregates 10 fitting results from vfits_i_1.R

  • Input: [protein].rda files in `data-r/``

  • Output: Rdata/vfit_[protein].Rdata

  • Scripts are run in high performance computing server environment where protein.name and nCores are provided as input arguments. (This script can be also run in a local computer as well.)

  • The folder code/vfits/Rdata/ is not uploaded in GitHub due to the file sizes. (Files can be recreated by running vfits.R script.)

code/vfits

  • R scripts: fits.R

  • fits.R contains a script fitting a single PU model at a specified py1 value.

code/enrich_comparison

  • R script: enr_test_fit.R

  • enr_test_fit.R contains a script fitting 10 CV PU and enrichment score models

  • Input: [protein].rda files in `data-r/``

  • Output: Rdata/enr_r_[protein].Rdata

  • Scripts are run in high performance computing server environment where protein.name, r, and nCores are provided as input arguments.

  • The folder code/enrich_comparison/Rdata/ is not uploaded in GitHub due to the file sizes

Related to results in MS

code/roc

  • R script: rocs.R

  • Input: vfits (code/vfits/Rdata)

  • Output: [protein].eps (for 10 proteins) (Figure 3 (a))

  • R script: aucs.R

  • Input: vfits (code/vfits/Rdata)

  • Output: aucs.csv (Figure 3 (b) for PU model)

code/enrich_comparison

  • R script: enr_test_summarize.R (code/enrich_comparison/Rdata)
  • Input: enr_[r]_[protein].Rdata (r=1,...,10)
  • Output: PU_enr_pvalue_diff.csv (Figure 3 (c)), comp_enr_diff.eps (Supplementary Figure 3 (b)),

code/coefficient_of_variation

  • R script: cv_fold.R
  • Input: vfits (code/vfits/Rdata)
  • Output: cv_density.png (Supplementary Figure 3 (c))

code/coefficient_selection_stability

  • R script: stability.R
  • Input: vfits (code/vfits/Rdata)
  • Output: avg_featsel_stability.eps (Supplementary Figure 3 (d))

code/py1_stability

  • R script: py1_stability.R
  • Input: vfits (code/vfits/Rdata)
  • Output: py1_stability.eps (Supplementary Figure 3 (e))

code/fits/Bgl3_HT

  • R script: pvalues_Bgl3_HT.R
  • Input: fit_Bgl3_HT_[r].Rdata (code/fits)
  • Output: top_10_muts.csv (Supplementary Figure 5 (b))