Authors: Joe Abbott, Keertana Krishnan, Guoyao Chen.
ExpressYeaself is an open source scientific software package that aims to quickly and accurately predict the contribution a promoter sequence has on the expression of genes in Saccharomyces cerevisiae (or 'Brewer's yeast ').
This will allow the costly and time-consuming trial-and-error processes in the development and synthesis of biotherapeutics to be streamlined. Our goal is to use machine learning and data mining to make recommendations on which promoter sequences are likely to contribute to high levels of gene expression, and which are not.
For further details on the scientific background of our project and back-end operation of our package, please see our use cases.
-
Raw data1 consisting of ~ 62 million sequences and their associated expression levels, can be processed in an automated, efficient, and highly tunable way, according to a large number of processing parameters.
-
Processed data is manipulated for input into our neural networks by one-hot encoding. This allows relationships between motifs within nucleotide sequences - and the effect they have on the expression level - to be learned on a deep level.
-
Three different models have been trained on this encoded data:
- 1-dimensional convolutional neural network (1DCNN)
- 1-dimensional locally connected network (1DLOCCON)
- Long-Short-Term Memory (LSTM), a type of recurrent neural network.
-
These trained models can then be used to make predictions on the extent to which each promoter sequence in a file will contribute to a gene's expression level.
This means a large input file of promoter sequences with potential for use in biotherapeutic drug design can be rapidly evaluated for their likelihood of being effective.
- We are currently in the process of developing some data mining tools to identify and extract so called magic motifs from our raw data.
- These are shorter nucleotide sequences that are present within complete promoter sequences that contribute to the highest expression levels.
- Identifying and extracting these will allow us to make recommendations on what motifs a promoter sequence should contain in order to result in a high expression level of the gene being promoted.
- Python 3.6.7 or later
- Conda version 4.6.8 or later
- GitHub
Execute the following commands in your computer's terminal application to install our package:
-
Clone the ExpressYeaself repository:
git clone https://github.com/yeastpro/ExpressYeaself.git -
Navigate into the repository:
cd ExpressYeaself -
Install our virtual environment:
conda env create -n environment.yml -
Enter the virtual environment:
conda activate yeast -
Download the raw data:
chmod +x download_data.sh && ./download_data.sh
Now you have installed our package and downloaded the raw data, you are ready to start using our features! You can use our interactive notebooks to take you through the process:
- Navigate into the directory containing our interactive guides:
cd expressyeaself/interaction/ - To start processing the data, use jupyter to open our first interactive notebook:
jupyter notebook 1_how_to_process_raw_data.ipynb & - Follow the instructions in the notebook, choose your parameters, and process the data.
- When you're done, save and exit the notebook.
- You can then start to encode your data and train your model:
jupyter notebook 2_how_to_train_model.ipynb &
ExpressYeaself (master)
|---doc
|---technology_reviews
|--1_sequencing_software_packages.md
|--2_neural_network_packages.md
|--timeline.md
|--use_cases.md
|---example
|---Abf1TATA_data
|--Abf1TATA_scaffold.txt
|---native_data
|--native_data.txt
|---pTpA_data
|--pTpA_scaffold.txt
|---processed_data
|--10000_from_20190610100252461788_homogeneous_deflanked_sequences_inserted_into_Abf1TATA_scaffold_with_exp_levels.txt.gz
|--10000_from_20190611170757656183_homogeneous_deflanked_sequences_with_exp_levels.txt.gz
|--10000_from_20190612130111781831_percentiles_els_binarized_homogeneous_deflanked_sequences_with_exp_levels.txt.gz
|--__init__.py
|--series_matrix_GSE104878-GPL17143.txt
|---expressyeaself
|---interaction
|--1_how_to_process_raw_data.ipynb
|--context.py
|---models
|---1d_cnn
|---saved_models
|--1d_cnn_classifier_onehot.hdf5
|--1d_cnn_parallel_onehot.hdf5
|--1d_cnn_sequential_onehot.hdf5
|--1D_CNN_builder.ipynb
|--context.py
|--native_sample.txt
|---1d_loccon
|--1d_locally_connected.ipynb
|--context.py
|--loc_con_1d.py
|---lstm
|--context.py
|--lstm_model_function.py
|---prediction_results
|--__init__.py
|---tests
|--__init__.py
|--context.py
|--test_build_promoter.py
|--test_encode_sequences.py
|--test_organize_data.py
|--test_process_data.py
|--test_utilities.py
|--__init__.py
|--build_promoter.py
|--construct_neural_net.py
|--encode_sequences.py
|--organize_data.py
|--process_data.py
|--utilities.py
|--version.py
|--.coveragerc
|--.gitignore
|--.travis.yml
|--LICENSE
|--README.md
|--download_data.sh
|--environment.yml
|--requirements.txt
|--runtests.sh
Any contributions to the project are warmly welcomed! If you discover any bugs, please report them in the issues section of this repository and we'll work to sort them out as soon as possible. If you have data that you think will be good to train our model on, please contact one of the authors.
1 Carl G. de Boer et al., Deciphering cis-regulatory logic with 100 million synthetic promoters, doi: http://dx.doi.org/10.1101/224907, 2017.
ExpressYeaself is licensed under the MIT license.
- Module not found errors:
- Make sure you're in our virtual environment!
- Re-enter it with:
conda activate yeast
- Permission denied errors when running shell scripts from terminal:
- You need to grant yourself access to execute the scripts.
- Do so with:
chmod +x <filename>.sh