Date: March 25th, 2021
Attendees: Bharath, Mufei, Ashwin, Yuanqi, Vignesh, Harsh
Summary: Bharath this week put out the 2.5.0 developer release of DeepChem. He still needs to write up developer notes. Which he hopes to do this weekend. Otherwise, Bharath just reviewed PRs.
Harsh is studying his bachelor of technology from IIT Dhanbad.
Mufei this week put up a PR adjusting the BBBP numbers. After checking the code from Michael, he found there was a bug.
Yuanqi was a bit busy the last few weeks and is now starting to catch up.Yuanqi asked if we want to include TDC datasets and models. Mufei pointed out an interesting talk about TDC at the DGL online meetup. Yuanqi suggested we put together a table tracking the datasets/models we want to benchmark for MoleculeNet 2.0.
Ashwin has been making himself more familiar with the library. He’s working to predict single step pathways with transformers. He hasn’t started implementing anything but hopes to get started.
Vignesh was busy this week but wasn’t able to do much. He’s hoping to get the megnet model up and running.
Bharath has some todos on his end to add the MoleculeNet benchmark table and to update the GSoC mentor post and wiki.
Date: March 19th, 2021
Attendees: Bharath, Peter, Andrew, David, Seyone
Summary: We had multiple new attendees so we started off with a round of introductions
Andrea is a scientist working on machine learning in biotech. He was most recently working on convolutional networks at Atomwise.
David is a chemist doing his Master’s degree in computational chemistry on molecular modeling, molecular dynamics and QM methods. He’s interested in learning to apply deepchem to computational biochemistry.
Pinhou is an undergrad from the university of liverpool. He’s doing research on the application of machine learning to small organic.
Peter was busy with other work this week.
Seyone was working on the fine-tuning script using HuggingFace. That simplified torch model implementation and script should hopefully be useful to setup the DeepChem ChemBERTa implementation. Seyone also worked on setting up the HuggingFace DeepChem profile, and how to transfer over the ChemBERTa models to the organization (https://huggingface.co/deepchem).
Stanley and Vignesh had a good discussion over email about integrating Ray Tune into DeepChem hyperparameter search. Stanley also came across https://github.com/kheyer/Genomic-ULMFiT.
Vignesh gave the same update.
Bharath gave a general update about GSoC, asking all students to send over their proposals for review.
Bharath asked about bumping to Python 3.8/3.9 for DeepChem 2.6.0. Peter said he didn’t know of any technical blockers and it should be feasible to do.
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.