Date: March 11th, 2021
Attendees: Bharath, Ashwin, Mufei, Vignesh, Peter, Dhairya
Summary: Bharath has had a slow week and has mostly been working reviewing PRs.
Vignesh created a PR for PAGTN in DGL-LifeSci. https://github.com/awslabs/dgl-lifesci/pull/139. Mufei has reviewed and merged it. Vignesh is planning to put up a wrapper for DeepChem and Mufei needs to make a minor release of DGL-LifeSci for the model. Vignesh asked about https://github.com/materialsvirtuallab/megnet and whether it would be interesting to wrap in DeepChem. Bharath agreed it would be interesting.
Dhairya is continuing to familiarize himself with DeepChem to work on the Julia version.
Peter this week has been working on editing the atomic contributions tutorial (PR)
We’ve been selected for Google summer of code! (See announcement).
Mufei asked about whether there were any ways to reproduce our old benchmark numbers from the original MoleculeNet paper. Bharath mentioned that the script was removed in 2.4.0, but was still available at https://github.com/deepchem/deepchem/blob/2.3.0/deepchem/molnet/run_benchmark.py in the 2.3.0 release.
Date: March 5th, 2021
Attendees: Bharath, Stanley, Seyone, Nathan
Summary: Bharath gave the same update as at the last meeting.
Stanley mentioned his dayjob has required him to use Ray, so he was interested in seeing whether we could add Ray hyperparameter tuning support to DeepChem. Bharath mentioned this might be a good Google Summer of Code project and Stanley agreed.
Seyone mentioned that this week on the ChemBERTa end, the ChemBERTa team chatted with HuggingFace to ask them to set up regression on HuggingFace. He’s also been working on setting up the DeepChem organization page on HuggingFace. Seyone asked everyone to make a HuggingFace account so he can add them on to the DeepChem organ page
We talked about https://arxiv.org/abs/2103.05247, an intriguing new paper which used transformer pretraining across a wide array of different applications.
Nathan this week fixed up the atomic convs and made a new tutorial (https://github.com/deepchem/deepchem/blob/master/examples/tutorials/14_Modeling_Protein_Ligand_Interactions_With_Atomic_Convolutions.ipynb) for the atomic convolutions. He also fixed a bug in RdkitGridFeaturizer PR.
Joining the DeepChem Developer Calls
As a quick reminder to anyone reading along, the DeepChem developer calls are open to the public! If you’re interested in attending either or both of the calls, please send an email to X.Y@gmail.com, where X=bharath, Y=ramsundar.