/PCSF_analysis

Primary LanguagePython

PCSF_analysis

PCSF workflow

  1. Parameter search
  2. Run forest.py on a parameter grid
  3. Visualize network features for each parameter set, and choose the best network
  4. Randomizations

Generate .yaml file

msgPath:        /nfs/apps/bin/msgsteiner9
pythPath:       /nfs/apps/python2.7/bin/python
forestPath:     /nfs/latdata/iamjli/alex/PCSF_analysis/bin/forest.py

projectPath:    /nfs/latdata/iamjli/alex/results/human_binned_weights/
terminals:      /nfs/latdata/iamjli/alex/data/toxicity_screen_human_prizes.tsv
interactome:    /nfs/latdata/iamjli/alex/data/interactome/iRefIndex_v13_MIScore_interactome.txt

# Parameter grid
w:      1,2,3,4
beta:   3,6,9,12
D:      8
mu:     3e-04,1e-03,3e-03,1e-02,3e-02

Pipeline

Run forest.py for all parameter sets, with optional parameter for specifying the yaml file:

python run_PCSF_param_sweep.py [yaml_file]

Run summarization scripts and visualize parameter grid results:

python PCSF_param_selection.py [yaml_file]

This also restructures the project directory. Results found in param_search/summary/. Choose best looking parameter set, and run randomization scripts:

python run_PCSF_randomizations.py W_3_BETA_3_D_7_mu_1e-05 [yaml_file]

Finally, summarize randomization results:

python randomizations2subnetworks.py

This script also performs community clustering (Louvain), finds GO terms for subclusters, and create output files for visualization in Cytpscape.

TODO:

  • script that fetches GO terms needs to handle empty responses
  • restructure randomization->summary steps so that more than one randomization can be run