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.
Public demo version of OmicSelector is available here.
Please note that this intance will reset and restart every Monday. All projects are purged every Monday! As this instance is shared with multiple users we also suggest not to upload sensitvie information to the demo platform.
Tailor the docker container image for your environment:
- 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
- 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.
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)
The recommended way of installing OmicSelector is presented above (Universal way). However, if you expirance difficulties, 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:
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).
Exemplary files for the analysis:
- Bugs and issues: https://github.com/kstawiski/OmicSelector/issues
- Contact with developers: Konrad Stawiski M.D. ([email protected], https://konsta.com.pl)
OmicApp is the framework utilizing OmicSelector to build complex Shiny applications. Please see https://github.com/kstawiski/OmicApp for more details.
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:
- Dr. Konrad Stawiski, M.D., Ph.D. ([email protected])
- Marcin Kaszkowiak, M.D.
- Damian Mikulski, M.D.
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)