Skip to content

Conversation

@mpvenkatesh
Copy link

The features previously implemented in the az repo have been incorporated into this cheminformatics repo. These include:

  • Being able to change the fingerprint length and radius through the UI
  • Generating a plot of the generated compounds
  • Naming the generated analogues better
  • A new way of generation that learns a direction to move in latent space by using a lasso regression to the specified property in the specified cluster, and then taking the specified number of steps in that direction from a specified number of compounds in the cluster.
  • Using a neural network with the specified architecture and activation functions to predict the specified property on a testing cluster after fitting to the training cluster
  • Performing similarity search on the GPU (and comparing it to the Tanimoto distances obtained by doing the same on the CPU through RDKit to show they are the same)

Rajesh Ilango and others added 26 commits September 7, 2021 14:39
With this change intermediate results are stored in sqlite database. All results
from generative models are stored in SQLite database. With this change, we
reduce the total number of requests to MegaMolbart gRPC service by 3/4th.
Additionally, training dataset is loaded into SQLite database and used while
computing Novelity metric. With this change, a request to check Novelity is
2ms.

Other changes include:
- Upgrade to rapids 2021.06
- Clean dockerfile to remove all workarrounds
- Remove the need for conda cuchem env inside the container.
- Ability to select docker image to build using launch script
- Add smile2embedding and embedding2smile to cddd
- Use hydra for benchmark configuration
RAPIDS 21.08 has fix for a bug filed for PCA on Multiple GPUs.

Other changes include:
 - 'GPU KMeans-UMAP - Single and Multiple GPUs' is now the default clustering
   workflow. This one has performance benefit compare to 'GPU KMeans-UMAP' while
   reclustering.

 - 'MegatronMolBART' is now the default generative workflow
 - Remove Apex installation while docker image creation.
Change directory names
Now docker-compose will be downloaded the first time the application
is launched.

Another changes is upgrade ansible script to install on ubuntu 20.04.
Also remote clear cache option while building docker image.
@mpvenkatesh mpvenkatesh requested a review from rilango February 28, 2022 04:19
Venkatesh Mysore added 2 commits February 27, 2022 20:48
@mpvenkatesh
Copy link
Author

@rilango , I believe my branch is ready for the merge with dev. Please let me know if you have any suggestions or comments.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants