Version Update: ImaGenev3.3v1:

Python script ImaGenev3_3v1.py is operational (use -h for help). Added to the main repo. New train.ini reflects sections for parameter-feed for new models and their respective grids. Note that this could be kept empty for defaults to govern.

Classification-based modelling techniques covered:

  1) Random Forest
  2) DecisionTree Classifier
  3) Multi-layer Perceptron (MLP) Classifier
  4) Support Vector Classifier (SVC)
  5) Logistic Regression

Regression-based modelling techniques covered:

  6) Decision Tree Regressor
  7) Linear Regression
  8) LinearModel
  9) multiTaskLinearModel
  10) LASSO
  11) multiTaskLASSO

Note: Docker container is not updated to the latest version, but older version 2. Stay tuned for updates!

Publication:

https://academic.oup.com/bioinformaticsadvances/article/2/1/vbac079/6820948

ImaGene

ImaGene provides researchers with a transparent tool for which to begin radiogenomic analysis and explore possible further directions in their research. A web platform is available here: https://www.imagene.pgxguide.org/index.php (Note: Please use the example data and label files as "Supplementary_Table_BC-Radiomic_features.csv" and "Supplementary_Table_BC-Gene_FPKM.csv" (available in the repo above) in "Input data features" and "Input label features" sections on ImaGene web platform. Kindly refer to the CLI section's Step 4's note below for the guidance on downloading these two input files successfully from the repo above. Kindly also view the "config_IBC_LR.ini" in the repo above for the example settings of values of several parameters that you may want to set/tune on ImaGene web platform during your experimental run thereon).

For analytical operations, ImaGene requires two input files, each in Comma-Separated Value (CSV) format – one containing imaging features (and their measurements) and another containing omics features for a set of tumor samples. The imaging features can be acquired from feature extraction software such as PyRadiomics, LIFEx, or RaCaT by processing the tumor images using the respective segmentation labels (Koçak et al. 2019; Pfaehler et al. 2019; van Griethuysen et al. 2017). The omics features can be acquired from studies conducted on tumor ROIs in biopsy samples processed in pathological laboratories, and may consist of data pertaining to gene expression, SV (including CNV), SNV, or DNA methylation scores.

ImaGene comprises of four modules: a) data pre-processing, b) correlation Analysis, c) machine learning (ML), and d) reporting (Figure 1). The code for the software has been written in Python, and it utilizes several libraries such as scikit-learn (Pedregosa et al. 2011), matplotlib, seaborn and importr along with custom functions written to follow a systematic approach to analyze and pin point meaningful associations between imaging and omics features.

Imp NOTE regarding input files:

1. Please do not leave spaces in the input file names: Example: "Supplementary Table BC Gene FPKM.csv". Kindly include either underscores or dashes in the names for the software to identify the files appropriately.
2. 'ID' column is the first column in the input file. It could be any characters or numbers as far as it distinguishes the samples from each other. 'ID' column is required in both the input files, i.e. "data" and "labels". Please make sure the order of the IDs are consistent between two files. Also, both files should have equal number of IDs in place. Otherwise, Software would yield error.

Silenting Correlations Threshold based filtering:

In case users wish to not filter correlations based on a correlation threshold, they could silent such filtration by setting the value of Correlation Threshold parameter to '-1'. Currently, the default value for that parameter is set to '0.5'.

New version 2:

Adding Imagene_v2_2.py (This python script runs has been packaged in the docker image-packet "Imagene_v2_2.tar.gz" available to download from the repo above). Steps to load docker package and run it successfully are described in the subsequent CLI-section below.

Changes in v2:

1. Automated permutations (n=20) conducted on labels that have AUC value > 0.9 and R-square > 0.25 from testing of models. From the results of the permutation tests, users could calculate p_value and 95% Confidence Intervals for a) AUCs for labels at various decision thresholds and b) R-squares for same labels.

2. Calculates R-squares using the sklearn.metrics.r2_score method. The results indeed match to the R-square calculated post ImaGene run using the formula presented in the manuscript i.e. R-square = 1 - v-square, where v-square = 1- ( RMSE(y)/Stdev(y) )^2, where 'RMSE/Stdev' is already indicated by ImaGene as one of its results for every label that is tested.

3. Calculating feature importances using "model.feature_importances_" method for DecisionTree model and "model.coef_" method for regression models. This indicates the importance (aka weightage) of each feature (i.e. imaging feature) when for example: a model is trained with imaging features to predict gene expressions of various genes (aka labels).

4. Added comments for function descriptions and other crucial steps (such as AUC calculations, feature importance reporting, etc.) within the respective functions.

Command Line interface (CLI) operation using docker

Step 1: Install Docker on an Ubuntu Machine using the following docker-installation documentation: https://docs.docker.com/engine/install/ubuntu/ For MacOS, use: https://docs.docker.com/desktop/mac/install/. Use docker using terminal app in Mac/Ubuntu thereafter.

Step 2: Load the docker container package Imagene_v2_2.tar.gz

    docker load < Imagene_v2_2.tar.gz

Step 3: Check the loaded image

    docker images
    
    Output:
    --------------------------------------------------------------------------------------------------
    REPOSITORY              TAG                 IMAGE ID            CREATED             SIZE
    shreysukhadia/imagene   2.2                 37297fbf58cb        12 days ago         1.3 GB
    --------------------------------------------------------------------------------------------------

Step 4: Run the docker image [Note: Download the data, label and config files as "Supplementary_Table_BC-Radiomic_features.csv", "Supplementary_Table_BC-Gene_FPKM.csv" and "config_IBC_LR.ini" respectively from the repo above. To download, please click on each file (in repo above) and then on the following page click on "Raw" button. Once the raw file opens up, go to your browsers menu (on top of your broswer) and then click on 'File', and then "Save page as" option therein, and save the file with its respective name (as it is). After downloading of all the three files completed, kindly execute the command as given below from the same working directory in linux/mac where you saved those three files]

    docker run -v "$(pwd)":/data shreysukhadia/imagene:2.2 Imagene_v2_2.py --data /data/Supplementary_Table_BC-Radiomic_features.csv --label /data/Supplementary_Table_BC-Gene_FPKM.csv --config /data/config_IBC_LR.ini > log_file 2>&1 &
  
    
    Output files: Same as indicated in supplementary (zip) files in the repo above and in the manuscript.

In case user wants to try including patient outcomes alongwith radiomic and/or omic features:

The user could put the patient outcomes as a column in a comma-separated values (csv) file and use that as a “label” file, and put radiomics or genomics (or radiomics and genomics combined) features as columns in another csv file and use that as a “data” file. The rows in each of these files have to represent the sample names and the values in the feature columns would represent the measures of the respective feature (for example: texture, shape, size or intensity measure for radiomics, gene expressions for genomics and patient outcome decision measures for clinical outcomes). This way the AI models would get trained and tested to predict patient clinical outcomes from radiomics or genomics or radio-genomics combined features.

REFERENCES:

Koçak, Burak, et al. (2019), 'Radiomics with artificial intelligence: A practical guide for beginners', Diagnostic and interventional radiology (Ankara, Turkey), 25 (6), 485-95.

Pedregosa, Fabian, et al. (2011), 'Scikit-learn: Machine learning in Python', Journal of machine learning research, 12, 2825-30.

Pfaehler, Elisabeth, et al. (2019), 'RaCaT: An open source and easy to use radiomics calculator tool', PLOS ONE, 14 (2), e0212223.

van Griethuysen, Joost J. M., et al. (2017), 'Computational Radiomics System to Decode the Radiographic Phenotype', Cancer research (Chicago, Ill.), 77 (21), E104-E07.

Please cite:

https://academic.oup.com/bioinformaticsadvances/article/2/1/vbac079/6820948

LICENSE:

MIT License

Copyright (c) 2021 Shrey Sukhadia

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.