Welcome to the github repoistory of iGEM Groningen 2021! Here you will find the code we used to incorporate the Automated Recommendation Tool into our project. The aim of using this tool was to make predictions on which combination of chassis, promoter, secretion peptide and alpha-amylase encoding gene would eventually result in a high alpha-amylase activity in the supernatant of the device in which all parts are combined. For an explanation on why there specific parts were chosen, please have a look at the Engineering page of our wiki.
In order to run the code on this repository, a licence for the ART needs to be obtained first. Instructions on how to do this can be found here. Once a licence has been obtained, follow the installation instructions and place the source code of art in the parent directory of this project. Have a look at the tutorials provided with the ART for an example of the correct project structure.
Once the ART has been properly installed, other dependencies can be installed using the following command:
pip install -r requirements.txt
The code is divided into three separate jupyter-notebooks, each representing one of the steps of the pipeline shown in figure 1
Step 1 is covered in Subsetting_combinatorial_library.ipynb. In this notebook, a balanced subset is created from a
combinatorial library. It produces a sample_list.csv where the measured activity is missing. Note that each part-variant
is assigned its own index. These indices need to be linked to the names of the actual parts using some sort of translation
dictionary. part_names.csv is used for this purpose.
Step 2 is covered in Preprocessing_v2.ipynb. This file uses sample_list.csv and raw absorbance measurements from
the Alpha-Amylase Assay to connect the samples with their activity measurement. The raw data used for this is located in
the /training_data folder. Each 96 well plate is given its own sub-folder where the name of the folder indicates the day at
which the measurements were taken. On the 5th of October, experiments were conducted on 2 96-well plates and hence there are two folders
starting with 5-10. More information about the raw measurements can be found in the README.txt located in each folder.
Note that the notebook needs to be executed once for each 96 well-plate.
In addition to extracting the activity, the natural logarithm of the activity is also determined and stored as
"Corrected activity". This is a preprocessing step needed for the ART. Both the "Activity" and "Corrected activity" of
the samples contained on the 96 well-plate are written into sample_list.csv.
Step 3 is covered in Performing_ART_analysis.ipynb. All the Machine Learning used for our project is covered in this
step. The produced model is thoroughly analysed to give some insights on how well the model will be able to generalise
to new data. Furthermore, the predicted corrected activity of the tested samples is saved to ART_out/filled_sample_list.csv.
Finally, 30 exploitative recommendations are saved to ART_out/exploit_reccomend.csv and 30 explorative recommendations are
saved to ART_out/explor_reccomend.csv.