Contains all code and data used for:
"A hybrid machine learning framework for functional annotation applied to mitochondrial glutathione metabolism and transport in cancers" - Luke Kennedy, Jagdeep K Sandhu, Mary-Ellen Harper, and Miroslava Cuperlovic-Culf.
💻📄Read the paper here
All code can be run using the packages specified below and data provided in data to reproduce the models and results of this work (unless specified otherwise in the main text). Placeholder strings for path directories in the code should be replaced with the path to where ever the data is located on the indivdual's system. Parameters for e.g., ML algorithm or model features can be specified in the code to replicate described models or produce novel models.
Make your own predictions: Example files and a brief tutorial are provided at the end under TUTORIAL to allow users to implement their own classifier and make predictions using this framework
Contains all python scripts used in data pre-processing, classifier model training, testing, and evaluation used for this work.
Contains all data files used in python code in this repository. 'info.txt' contains specific information for each data file.
Contains all python scripts used for collecting required data (FASTA sequences) for calling DeepGOWeb API, and evaluation of DeepGO annotations and RF classifier annotations.
Contains all python scripts used for structural analysis of SLC25 proteins, and for generating corresponding output figures. CAVER was implemented via plugin through PyMol GUI and results are summarized in SLC25_tunnelRes.pkl, generated by mGSH_CAVER_resis.py from CAVER plugin.
The following version of Python and Python packages are used in the code provided here. All code was implemented on Windows with an installation time of 15 minutes.
Python --3.8.16
- numpy --1.23.5
- pandas --1.5.2
- scikit-learn --1.2.1
- scipy --1.10.0
- matplotlib --3.6.3
- matplotlib-venn --0.11.9
- seaborn --0.12.2
- pymol --2.4.1
- pymol-psico --4.2.2
- requests --2.28.1
Installation information:
pip install numpy
pip install pandas
pip install scikit-learn
pip install scipy
pip install matplotlib
pip install matplotlib-venn
pip install seaborn
conda install -c schrodinger pymol
conda install -c speleo3 pymol-psico
pip install requests
conda install openpyxl <- Only necessary for using example generic code "run_model.py"
conda/pip install foo=1.2.3 <- For installing specific versions
For individuals interested in applying this classifier framework to their topic or function of interest, we have provided a generic run_model.py file which can be run in the command prompt in environments which contain a Python installation with the required packages, for example:
python run_model.py --help
To get information about the various requirements to run the code.
To run a simple classifier for glutathione metabolism based on a subset of CCLE transcriptomics principal components, unzip ccleTranscriptomics_PCA.csv in the data folder and, with that and the exampleGSH_genes.csv file, enter the following into the command prompt after changing to the directory where these files are located to generate predictions:
python run_model.py --data-path "data" --labeled-genes "example_GSHgenes.csv" --feature-data "ccleTranscriptomics_PCA.csv" --select-features 5 --iterations 10 --cpu-cores 2
The outputs contain test set results found in the "Results" csv file, and classification probabilites for unlabeld samples are found in the "unknownResults" csv file under "Average predicted label".
In this example, these classifiers predict annotations for gene ontology terms of interest based on CCLE transcriptomics data, however, this framework and run_model.py can be used in many applications, such as classifying metabolites based on a metabolomics dataset and a dataset of metabolite biochemical properties. The only requirements are that rows and columns correspond to samples and features/observations, respectively, for use in classifiers and that sample identifiers are consistent across files (e.g., Ensembl gene/transcript IDs, UniProt IDs, Metabolite names).
Simply entering the above command and replacing labeled-genes, feature-data, and possible other-features with the feature and label data of interest, meeting the aforementioned requirements, is sufficient for generating a novel classifier and predctions.
We hope this code provides an easy implementation of our classification framework which can be used to generate hypotheses for biology! If you encounter and issues or would like to make a suggestion, please open an issue so it can be addressed.