I got problems with the package "parse" since even though was installed couldn't run the program without getting "no module named parse". I think it has to be related with an incompatibility between python's versions (2.7 / 3). To solve this issue "manually", just rewrite few lines of different py-files in ../chembl_websource_client/ (e.g. replacing "from urllib.parse import urlparse" to "from urlparse import urlparse") so that could be executed with python2.7

Travis Error

modtox/data/dude.py:15: in
from chembl_webresource_client.unichem import unichem_client as unichem
../../../conda/envs/tenv/lib/python2.7/site-packages/chembl_webresource_client/unichem.py:5: in
from urllib.parse import urlencode
E ImportError: No module named parse

Travis Solution

In modtox/data/dude.py:15:

try:
from urllib.parse import urlparse
except ImportError:
from urlparse import urlparse

We should check the performances over an external dataset.

Add three more components in UMAP & PCA plot which give a meaning of how different are the 2D points in further dimensions.

Wrong indexing when reading from glide csv: molecules stored in different order in csv and sdf file.

Issue

Use --dude flag to build model from scratch. Get to know the code. See input/output and understand the workflow

Dataset

http://dude.docking.org/targets/cp2c9

Found AUC-ROC weird values. Check sklearn function.

Mistaken samples name are not right becoming confusing when debugging. We should have a look

We should implement an applicability domain method to discard outliers when training and non-predictable instances.

We could include the ones below to the models uncertainty:

1- Applicability domain

2- Drop out

We should improve installation : fix conflict between packages ?

Issue

Right now xglide is supported but not SMP. Work together with Dani and Lucia to make it work there

Description.......

Start building a base model for structural alert. Start from the CNN we have.

Compute the thresholds for all the training molecules and the euclidean distances from each of the the test-set to the the train-set and count how many thresholds include each one of those molecules. Finally order molecules and write a txt file.

Current method for plotting confusion matrix is only working well with python2.7. We should explore new solutions for python3.

Dude.py script is a little messed up, so should be improved.

Check all available CYPs and the datasets

Delete some useless files and cover the others.

Issue

Build model CYP11A1 from chembl to understand how it works and what is the output

Dataset

https://www.ebi.ac.uk/chembl/target_report_card/CHEMBL3522/