IDG Dream challenge

You can install the conda environment by running :

conda env create -n idg_dream -f environment.yml

Then activate it :

conda activate idg_dream

Depending if you have cuda enabled or not :

pip install -r cpu_requirements.txt

or

pip install -r cuda_9_requirements.txt

if you have another cuda version, check the pytorch documentation.

In order to use machine learning algorithms we will need additional information about the compounds and proteins.

• For compound information I decided to use CHEMBL as it is the provided compound_id in the dataset.
• For protein information I used Uniprot for the same reason

For both sources of information I decided to download the databases and to restore them to a local postgres database dedicated to the project.

To start the postgres container, run :

IDG_DREAM_DB_PORT=5432 docker-compose -f idg_dream_db/docker-compose.yml up -d

Note that IDG_DREAM_DB_PORT may be any port you like.

Then to restore the data you will need to run through the following procedure in order.

You can download the postgres dump from :

download chembl database

Extract the files and upload the content of the previous download to the database by running :

cat PATH_TO_CHEMBL_DUMP | docker exec -i idg-dream-db pg_restore -O --username=idg_dream -d idg_dream

This takes some time.

For now we will only use the protein sequence thus requiring only the fasta dump. You can get it from :

download uniprot

Extract the file and run :

PYTHONPATH=PATH_TO_PROJECT:$PYTHONPATH python bin/import_uniprot.py FASTA_PATH

Again, the port and host may be specified as options.

This can take some time too.

The data used to create the training set is issued from the DTC website :

download dtc

In order to understand the details of the process of the training set creation you can look at the ipython notebook data_analysis.ipynb.

To create the table containing the training set, you can use the following script :

PYTHONPATH=PATH_TO_PROJECT:$PYTHONPATH python bin/create_training_set.py DTC_DATA_PATH

Again you can provide a port and host in the database is on a remote server.

In order to run tests, you will need a postgres image running, you can use the given docker-compose file :

docker-compose -f idg_dream_db/docker-compose.test.yml up -d

and run :

python -m unittest discover tests/