deepchain-apps 0.0.17

Define a personnal scorer for the user of DeepChain.bio

deepchain-apps : Package for creating a personnal app to deploy on the DeepChain platform

Getting started with Scorer

This template provide an example of application that you can submit. The final app must have the following architecture:

• my_application
  • src/
    • scorer.py
    • Optionnal : requirements.txt (for extra packages)
  • checkpoint/
    • Optionnal : model.[h5/pt]

The main scorer class must be named ’Scorer’

CLI

The CLI provides 3 main commands:

• login : you need to supply the token provide on the plateform (PAT: personnal access token).

deepchain login

• create : create a folder with a template scorer file

deepchain create my_application

• deploy : the code and checkpoint are deployed on the plateform, you can select your app in the interface on the plateform.
  • with checkpoint upload

    deepchain deploy my_application --checkpoint

  • Only the code

    deepchain deploy my_application

The application will be deploy in DeepChain plateform.

Embedding

Some embeddings are provided in the TransformersApp module

from deepchainapps.components import TransformersApp

The model are furnished, but not mandatory, if you want to make an embedding of your protein sequence. Only the ESM (evolutionary scale modeling) model is provided, with different architecture. Here for some full details of the architecture (https://github.com/facebookresearch/esm)

• 'esm1_t6_43M_UR50S'
• 'esm1_t12_85M_UR50S'
• 'esm_msa1_t12_100M_UR50S'
• 'esm1b_t33_650M_UR50S'
• 'esm1_t34_670M_UR100'
• 'esm1_t34_670M_UR50D'
• 'esm1_t34_670M_UR50S'

!! The embedding will run on a GPU on the plateform. But for a testing phase on your personal computer (CPU), you should choose the smaller architecture.