/live_cell_reporter_imaging

Primary LanguageJupyter Notebook

Live Cell Reporter Imaging

DOI

Project in Dr. Gordon Mills Lab at Oregon Health and Science University led by Dr. Samuel Tsang, data analysis by Nate Evans (evansna@ohsu.edu).

To run analysis...

  1. Download data from one drive. Email evansna@ohsu.edu for access.

    • if you are extracting the relevant files from Samuel's file structure, then use ./src/HER2_extract_data2.sh

  2. Unpack zip files ./data/

  3. build and activate the conda environment

# build conda env if necessary 
conda env create -f environment.yml 

# activate env 
conda activate lc_reporter
  1. Run analysis bash script (Note: this can take quite a while; 8+ hours):
$ cd ./src/
$ ./HER2_sensitivity_runs.sh

See HER2_sensitivity_runs.sh for input/output paths/

  1. Aggregate the results by:
$ python agg_results.py --input ../output/ --out ../output/

  1. Use the sensitivity_analysis.ipynb notebook to visualize sensitivity analysisresults.

  2. Finalize results (aggregate across sensitivity analysis runs) by:

$ python finalize_results.py --input ../output/ --output ../output/

This script aggregates all results by (cell line, treatment, mutant, batch) and merges the batch effect flag.

Will save 5 files:

- final_results.csv -- (all results)
- final_results-sorted-EFM192A;NERATINIB.csv
- final_results-sorted-EFM192A;TRASTUZUMAB.csv
- final_results-sorted-SKBR3;NERATINIB.csv
- final_results-sorted-SKBR3;TRASTUZUMAB.csv

Results are sorted by the mean probability of resistance across all sensitivity analysis runs.

Data Dictionary

final_results....csv

applies to:

- final_results.csv 
- final_results-sorted-EFM192A;NERATINIB.csv
- final_results-sorted-EFM192A;TRASTUZUMAB.csv
- final_results-sorted-SKBR3;NERATINIB.csv
- final_results-sorted-SKBR3;TRASTUZUMAB.csv

cell_line: SKBR3 or EFM192A

treatment: neratinib or trastuzumab

mutant: specific mutation induced in the HER2 gene

batch: dataset source

mean: average probability of resistance, aggregated across all sensitivity analysis runs

std: standard deviation of probability of resistance, aggregated across all sensitivity analysis runs

q05: 5th percentile of probability of resistance, aggregated across all sensitivity analysis runs

q95: 95th percentile of probability of resistance, aggregated across all sensitivity analysis runs

min: minimum value of probability of resistance, aggregated across all sensitivity analysis runs

max: maximum value of probability of resistance, aggregated across all sensitivity analysis runs

any_flag: proportion of batch effect flags across all sensitivity analysis runs

mutant_resistance_results.csv

pc1: First principle component, latent representation of a cell line's response to drug (many cell's reporter values over time)

pc2: Second principle component

treatment: The drug treatment used in a given condition.

mutant: The cell lines induced HER2 mutation used in a given condition

batch: The experimental batch identifier. Each batch has it's own set of sensitive and resistant controls.

cell_count: The number of cells measured and recorded for each mutant in a given batch and treatment. Note, there will be a unique cell count for each unique (batch,mutant,treatment) observation.

prob_res: Predicted probability of resistance [as defined by the resitant controls].

prob_sens: Predicted probability of sensitivty [as defined by the sensitive controls (WT, treated)].

call: [sens, res] Most probable call (binary)

run_id: Unique analysis identifier; necessary for distinguishing results from different analysis with different hyperparameters in the sensitivity analysis.

cell_line: The cell line used in a given condition (SKBR3, EFM193A)

drug_check: QC check that merge performed properly.

pc1_coef: Batch PCx adjustment; If batch correction is used, then $pcx = pcx_uncor - pcx_coef$. Nan if batch correction not used.

pc1_pval: Batch effect significance, as measured by ANOVA [Fit regression using controls: PCx ~ batch].

pc2_coef: Batch PCx adjustment; If batch correction is used, then $pcx = pcx_uncor - pcx_coef$. Nan if batch correction not used.

pc2_pval: Batch effect significance, as measured by ANOVA [Fit regression using controls: PCx ~ batch].

pc1_uncor: The uncorrected PCx values. Nan if batch correction not used.

pc2_uncor: The uncorrected PCx values. Nan if batch correction not used.

experiment_run_results.csv

accuracy(train): The classifier accuracy on training data (sensitive and negative controls, treated); Describes how separable the controls are.

pc1_var: Amount of variance explained by PCx

pc2_var: Amount of variance explained by PCx

kmeans_inertia: The inertia (within cluster sum of squares) of the time-series k-means fit.

res_line: Resistant mutant used in a given analysis; Note: "line" is a misnomer, it does not refer to SKBR3 or EFM192A.

sens_line: Sensitive mutant used in a given analysis; Note: "line" is a misnomer, it does not refer to SKBR3 or EFM192A.

drug: The drug used, synonymous with treatment in other results .csvs.

nclus: The number of clusters used in the time-series k-means fit.

resample_sz: The time series resample size used.

load: [normalized, raw] The data form used in the analysis. TODO: Link to description of normalization.

burnin: The burn-in used in the analysis - this is the number of measured points removed from the beginning of each cell lines time series. Intended to alieve batch effects and remove the common high-to-low reporter behavior observed at the beggining of an experiment.

batch_corrected: [True, False] whether PC-space batch correction was applied.

run_id: Unique analysis identifier; necessary for distinguishing results from different analysis with different hyperparameters in the sensitivity analysis.

cell_line: [SKBR3, EFM192A] The cell line used for a given condition.

drug_check: QC check that merge performed properly.

batch_effect_results.csv

batch: The experimental batch identifier. Each batch has it's own set of sensitive and resistant controls.

pc1_coef: Batch PCx adjustment; Fit regression using controls: PCx ~ batch.

pc1_pval: Batch PCx significance via ANOVA; Fit regression using controls: PCx ~ batch.

pc2_coef: Batch PCx adjustment; Fit regression using controls: PCx ~ batch.

pc2_pval: Batch PCx significance via ANOVA; Fit regression using controls: PCx ~ batch.

run_id: Unique analysis identifier; necessary for distinguishing results from different analysis with different hyperparameters in the sensitivity analysis.