ersilia-os / eos30gr

GNU General Public License v3.0
0 stars 1 forks source link

Retrain Model eos30gr with a multi-task deep neural network (DNN) without mol2vec #10

Closed Malikbadmus closed 4 months ago

Malikbadmus commented 4 months ago

Description

In deepchem the MultitaskClassifier model has 6 acceptable featurizer:

CircularFIngerprint was selected for this model.

Further Discussion here #8

Changes made

Model retrained on the original author's method but without mol2vec + MOE featuization.

Status

The model utilizes a PyTorch backend,I faced some difficulties in getting PyTorch to install on Ersilia, causing compatibility issues and dependencies conflict, finally resolved this by making use of the conda-forge channel.