ProtoGain

WORK STILL IN PROGRESS

In this repository, you may find a PyTorch implementation of Generative Adversarial Imputation Networks (GAIN) [1] for imputing missing iBAQ values in proteomics datasets.

Table of Contents

• Installation
• How to Use
• Demo
• References

Installation

1. Clone this repository: git clone https://github.com/QuantitativeBiology/ProtoGain/
2. Create a Python environment: conda create -n proto python=3.10 if you have conda installed
3. Activate the previously created environment: conda activate proto
4. Install the necessary packages: pip install -r requirements.txt

How to Use

If you just want to impute a general dataset, the most straightforward and simplest way to run ProtoGain is to run: python protogain.py -i /path/to/file_to_impute.csv Running in this manner will result in two separate training phases.

1. Evaluation run: In this run a percentage of the values (10% by default) are concealed during the training phase and then the dataset is imputed. The RMSE is calculated with those hidden values as targets and at the end of the training phase a test_imputed.csv file will be created containing the original hidden values and the resulting imputation, this way you can have an estimation of the imputation accuracy.

2. Imputation run: Then a proper training phase takes place using the entire dataset. An imputed.csv file will be created containing the imputed dataset.

However, there are a few arguments which you may want to change. You can do this using a parameters.json file (you may find an example in ProtoGain/breast/parameters.json) or you can choose them directly in the command line.

Run with a parameters.json file: python protogain.py --parameters /path/to/parameters.json
Run with command line arguments: python protogain.py -i /path/to/file_to_impute.csv -o imputed_name --ofolder ./results/ --it 2001

Arguments:

-i: Path to file to impute
-o: Name of imputed file
--ofolder: Path to the output folder
--it: Number of iterations to train the model
--miss: The percentage of values to be concealed during the evaluation run (from 0 to 1)
--outall: Set this argument to 1 if you want to output every metric
--override: Set this argument to 1 if you want to delete the previously created files when writing the new output

If you want to test the efficacy of the code you may give a reference file containing a complete version of the dataset (without missing values): python protogain.py -i /path/to/file_to_impute.csv --ref /path/to/complete_dataset.csv

Running this way will calculate the RMSE of the imputation in relation to the complete dataset.

Demo

In this repository you may find a folder named breast, inside it you have a breast cancer diagnostic dataset [2] which you may use to try out the code.

breast.csv: complete dataset
breastMissing_20.csv: the same dataset but with 20% of its values taken out

To simply impute breastMissing_20.csv run: python protogain.py -i ./breast/breastMissing_20.csv
If you want to compare the imputation with the original dataset run: python protogain.py -i ./breast/breastMissing_20.csv --ref ./breast/breast.csv or python protogain.py --parameters ./breast/parameters.json

If you want to go deep in the analysis of every metric you either set --outall to 1 or you run the code in an IPython console, this way you can access every variable you want in the metrics object, e.g. metrics.loss_D.

References

[1] J. Yoon, J. Jordon & M. van der Schaar (2018). GAIN: Missing Data Imputation using Generative Adversarial Nets
[2] https://archive.ics.uci.edu/dataset/17/breast+cancer+wisconsin+diagnostic