tandem-dimple-app

Webapp for Tandem-dimple project

Installation

First, clone the repo and the submodule (Tandem-Dimple)

git clone <path_to_git>
git submodule update --init

We need to first have the environment to download the database. This is used by Loci's downloading script.

conda env create -f environment_for_downloading_database.yml

Then, we can download the required database for Loci's inference code

conda activate tandem

# In ./inference/tandem/
bash scripts/download_pfam.sh data/pfamdb # 1.5G, ~1.5m
bash scripts/download_consurf_db.sh data/consurf/db # 2.5G, ~2m
# Please skip this database for now
# We will download this database later
bash scripts/download_uniref90.sh data/consurf # 90G, ~127m

Then, we can build and run the docker containers.

# Build and run all the containers
docker compose up -d --build

# Stop all containers
docker compose down

We can check the frontend at http://0.0.0.0:7860/

Database

• user_input

  Its status would go from pending -> processing -> finished

What each docker does?

• gradio_app

  Frontend for the website, exposed on port 7860.

  Would check for and update the result from the database automatically, for every 3 seconds.

• mongodb

  Database

  container platform: linux/amd64 or linux/arm64 both are OK, very stable. But don't use linux/amd64 on Apple Silicon, since Rosetta does not have CPU AVX support.

• worker

  Automatically fetch user input one-by-one, whose "status" are "pending".

  Use atomically lock jobs, whenever an input is fetched, its "status" changes from "pending" -> "processing"

  Jsonify the input and send to the inference container docker, through HTTP API.

  Write the input with the results, back to database, "status" changed from "processing" to "finished".

• inference

  Perform feature processing and model inference.

  The inference docker build the correct environment needed for Loci's inference code.

  Container platform can only use linux/amd64, and cannot run on Apple Silicon, since TensorFlow needs CPU AVX support.

  Serve a /infer API through flask, on internal port 5000.

  Use git submodule to link to Loci's inference git repo.

  Use adapter.py to import Loci's inference functions.

Reach the website outside, but now Typhoon 140.114.97.194

http://140.114.97.192:7860

Reach the website outside, but now A100 140.114.97.192

http://140.114.97.192:7860 http://140.114.97.207:7860