Skip to content

OmicSelector - automatic feature selection and deep learning modelling for omic experiments.

OmicSelector is the environment, docker-based application and R package for biomarker signiture selection (feature selection) & deep learning diagnostic tool development from high-throughput high-throughput omics experiments and other multidimensional datasets. It was initially developed for miRNA-seq (small RNA, smRNA-seq; hence the previous name was miRNAselector), RNA-seq and qPCR, but can be applied for every problem where numeric features should be selected to counteract overfitting of the models. Using our tool, you can choose features, like miRNAs, with the most significant diagnostic potential (based on the results of miRNA-seq, for validation in qPCR experiments). It can also develop the best deep learning model for your signature, as well as be an IDE for your more complex data mining project (contains R Studio, Jupyter notebooks and VS Code.. all integrated in one!).

The primary purpose of OmicSelector is to provide you with the set of candidate features (biomarkers) for further validation of biomarker study from, e.g., high-throughput experiments. The package performs feature selection first. In the next step, the sets of features are tested in the process called “benchmarking”. In benchmarking, we try all of those biomarkers’ sets using various data-mining (machine learning) methods. Based on the average performance of groups in cross-validation or holdout-validation (testing on the test set and/or validation set), we can suggest which of the signatures (set of features) have the tremendous potential for further validation.

As the feautres are selected, OmicSelector can perform advanced modeling of deep feedforward neural networks with and without autoencoders. The best network is developed using comperhensive grid search of optimal hyperparameters. This section works with Tensorflow (via Keras), so the computations can be GPU-accelerated! The best network can be easily implemented in clinical practice using our interactive application.

Try it out

Biostat.umed.pl demo

Link: https://biostat.umed.pl/OmicSelector/demo/

Please note that this intance will reset and restart every Monday. All projects are purged every Monday!

Please note that uploading real data to this instance is not safe. You’re data and analysis files will be accessible by anyone (public). The great potential for customization and extension of the environment comes with some security flaws (e.g., access to files via shell or VS Code), so we highly discourage the users from using this instance for real-life projects.

Ainize demo

Run on Ainize

Link: https://master-omic-selector-kstawiski.endpoint.ainize.ai/

Please note that uploading real data to this instance is not safe. You’re data and analysis files will be accessible by anyone (public). The great potential for customization and extension of the environment comes with some security flaws (e.g., access to files via shell or VS Code), so we highly discourage the users from using this instance for real-life projects. If you wish to go deeper, i.e. messing with files as root, you should fork the repo. You can get your own free and working OmicSelector using Ainize. [Note: if you wish to build your own docker change FROM kstawiski/omicselector-gpu in Dockerfile]

Installation with GUI

Tailor the docker container image for your environment:

  1. GPU-based, using Nvidia CUDA: kstawiski/omicselector-gpu 
docker run --name OmicSelector --restart always -d -p 28888:80 --gpus all -v $(pwd)/:/OmicSelector/host/ kstawiski/omicselector-gpu
  1. CPU-based: kstawiski/omicselector 
docker run --name OmicSelector --restart always -d -p 28888:80 -v $(pwd)/:/OmicSelector/host/ kstawiski/omicselector

As the docker image updates itself, it may take few minutes for the app to be operational. You can check logs using docker logs OmicSelector. The GUI (web-based user interface) is accessible via http://your-host-ip:28888/. If you use the command above, Omicselector will bind your working directory as /OmicSelector/host/.

Pearls:

  • Docker version contains a web-based GUI allowing for easy implementation of the pipeline.
  • Advanced features allow running Jupyter-based notebooks, allowing for modification
  • Contains Jupyter-notebook-based tutorial for learning and easy execution of R package.
  • For the Docker-based version, we assure the correct functionality. Docker container is based on configured ubuntu.

If you have a compatible GPU you can consider changing tensorflow to tensorflow-gpu in conda install command. ## Installation without GUI (just package)

Universal

Setup the package in your own R enviroment. You need to have your system prepared (prerequirements installed).

library("devtools") # if not installed, install via install.packages('devtools')
source_url("https://raw.githubusercontent.com/kstawiski/OmicSelector/master/vignettes/setup.R")
install_github("kstawiski/OmicSelector", force = T)
library(keras)
install_keras()
library(OmicSelector)

Linux/macOS using conda-pack

This approach uses conda-pack to install the conda enviornment developed using Github workflows. Conda should be installed to use this approach.

conda install conda-pack # install conda-pack
wget https://github.com/kstawiski/OmicSelector/releases/download/release/OmicSelector_conda_pack.tar.gz.partaa
wget https://github.com/kstawiski/OmicSelector/releases/download/release/OmicSelector_conda_pack.tar.gz.partab
wget https://github.com/kstawiski/OmicSelector/releases/download/release/OmicSelector_conda_pack.tar.gz.partac
cat OmicSelector_conda_pack.tar.gz.part* > OmicSelector_conda_pack.tar.gz
rm OmicSelector_conda_pack.tar.gz.part*
mkdir -p OmicSelectorEnv
tar -xvzf OmicSelector_conda_pack.tar.gz -C OmicSelectorEnv
rm OmicSelector_conda_pack.tar.gz
source OmicSelectorEnv/bin/activate
conda-unpack

Alternatively, you can download already merged file from: https://deploy.konsta.com.pl/OmicSelector/release/OmicSelector_conda_pack.tar.gz

Windows using conda-pack

Windows conda-pack is available for download here: https://deploy.konsta.com.pl/OmicSelector-win/OmicSelector_conda_pack_win.tar.gz

Copy the OmicSelector_conda_pack_win.tar.gz from the source environment to the target environment. Unpack the tar.gz environment file to where you want. On Windows, you can do this with 7-zip. It’ll take one step to decompress the gzip and another step to unpack the tar. Be sure you’re using standard Command Prompt (cmd.exe) on Windows, not Anaconda Prompt or PowerShell; those shells don’t correctly add the right files to PATH. In the next step run following commands in the unpacked directory:

.\Scripts\activate.bat
.\Scripts\conda-unpack.exe

Linux/macOS using conda (own build)

  1. Installing the package in your own Anaconda environment:

Use, e.g., conda create -n OmicSelector and conda activate OmicSelector to set up your environment. Please note the this will work only when running on Linux (Ubuntu) OS or macOS.

conda install --channel "conda-forge" --channel "anaconda" --channel "r" tensorflow keras jupyter jupytext numpy pandas r r-devtools r-rgl r-rjava r-mnormt r-purrrogress r-xml gxx_linux-64 libxml2 pandoc r-rjava r-magick opencv pkgconfig gfortran_linux-64
echo "options(repos=structure(c(CRAN='http://cran.r-project.org')))" >> ~/.Rprofile
Rscript -e 'update.packages(ask = F); install.packages(c("devtools","remotes")); remotes::install_cran("pkgdown");'
Rscript -e 'devtools::source_url("https://raw.githubusercontent.com/kstawiski/OmicSelector/master/vignettes/setup.R")'

If you have a compatible GPU, you can consider changing tensorflow to tensorflow-gpu in conda install command.

Windows OS

You can download our Windows-based R environment from here: https://studumedlodz-my.sharepoint.com/:u:/g/personal/btm_office365_umed_pl/EQUihquz915JoVhsQQShcnoBZaukMkwd3MnC1LER0iORNw?e=W6KEyu

After unpacking, if you wish to use our enviorment, please consider setting our R version in your R Studio installation:

Tutorials

Video tutorial

Video tutorial: https://www.youtube.com/watch?v=dKUdINEcOjk

This tutorial shows how OmicSelector’ GUI works and how to perform (without programming knowledge):

  • Feature selection
  • Benchmarking (selecting best set of variables based on the performance of data-mining models)
  • Deep learning model development (feedforward neural network up to 3 hidden layers and with/without autoencoders; grid search of hyperparameters)
  • Exploratory analysis (differential expression using t-test, imputation of missing data using predictive mean matching, correcting the batch effect using ComBat, generating heatmaps and volcano plots).

Resources

Exemplary files for the analysis:

Development

DockerR package (Linux) R package (Windows)

Build with OmicSelector

OmicApp is the framework utilizing OmicSelector to build complex Shiny applications. Please see https://github.com/kstawiski/OmicApp for more details. ## Footnote

Citation:

Stawiski K, Kaszkowiak M, Mikulski D, Hogendorf P, Durczynski A, Strzelczyk J, et al. OmicSelector: automatic feature selection and deep learning modeling for omic experiments. bioRxiv. 2022. p. 2022.06.01.494299. doi: https://doi.org/10.1101/2022.06.01.494299

Available on bioRxiv as preprint.

Authors:

Supervised by: prof. Wojciech Fendler, M.D., Ph.D. 

For any troubleshooting use https://github.com/kstawiski/OmicSelector/issues.

Department of Biostatistics and Translational Medicine, Medical University of Lodz, Poland (https://biostat.umed.pl)