DeepChem Minutes 4/22/2021

DeepChem Minutes

India/Asia/Pacific Call

Date: April 22nd, 2021
Attendees: Bharath, Mufei, Ashwin
Summary: Bharath worked with Alana on the website and reviewed some PRs.

Mufei has been busy over the last few weeks but now has a bit more time. Mufei is working with Vignesh to enable the PagTN model. Mufei will also try to work on benchmarking for MoleculeNet.

Ashwin has continued working on the retrosynthesis engine. The URL for the USPTO dataset is loader is broken. The current URL is no longer functional and we need to get different datasets located and uploaded onto DeepChem. Ashwin is also having some difficulty with the SmilesTokenizer

Bharath recommended that Ashwin try to implement a baseline model. Ashwin asked what baseline should be used here. Bharath suggested using the SeqToSeq model. Ashwin suggested that it might be useful to use a basic transformer from pytorch as well.

Americas/Europe/Africa/Middle East

India/Asia/Pacific Call

Date: April 23rd, 2021
Attendees: Bharath, Peter, Sanjay, Stanley, Vignesh, Seyone, James
Summary: Sanjay is a first time attendee. Sanjay is a product manager at Benchling who’s working on chemistry products there.

Bharath gave the same update as last week.

Peter has been updating the code for the book. One of the chapters in the book hadn’t been included in the repo and required larger updates.

Stanley’s team was giving a try to Anyscale’s Beta. Stanley was asking how to incorporate Ray into DeepChem. Stanley asked about where we want to use tools like Ray in DeepChem, and Bharath said featurization and transformation might be useful.

Vignesh has fixed a bug in DeepChem (PR) and is working on PagTNModel. Vignesh asked whether the Ray usage would be for training or featurization and Bharath said probably featurization first.

Seyone has been working on using the ChemBERTa sequence classification models to train on MoleculeNet multitask issues. Seyone is trying to figure out how there might be a good design since we have to inherit from both DeepChem and HuggingFace models.

David has been looking to add uniprot data to DeepChem. David asked what uniprot data we want to add to DeepChem, just the raw sequence or do we want annotations or other data?

James hasn’t been on the call for a bit so joined in to just say hi to folks.

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.

1 Like