https://github.com/xrsrke/progen

Generating new proteins using language models
https://github.com/xrsrke/progen

Last synced: about 2 months ago
JSON representation

Generating new proteins using language models

• Host: GitHub
• URL: https://github.com/xrsrke/progen
• Owner: xrsrke
• License: mit
• Created: 2023-02-03T06:02:30.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-02-25T03:49:38.000Z (over 2 years ago)
• Last Synced: 2025-03-27T11:13:55.832Z (2 months ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 3.93 MB
• Stars: 4
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
ProGen - 🚧 WORK IN PROGRESS 🚧

================

![image.png](index_files/figure-commonmark/824a0f6d-1-image.png)

Paper: [Large language models generate functional protein sequences

across diverse

families](https://www.nature.com/articles/s41587-022-01618-2)

I am currently working towards this paper. Check out my learning

progress here: https://twitter.com/xariusrke/status/1621403313651728386

### TODO

- \[DONE\] fast2dict

- Extract multiple properties of a protein sequence

- Tokenize the control tag

- Generate training data from dataset

### Questions

**Tokenizing** - Does the control tag use the same tokenizer as protein

sequence? - A control tag is represented by a token or multiple tokens?

If multiple tokens, then conditioned them all? - How to separate each

control tag? - Can the model predict the next control tag?

**Training**

### Resources

I implemented `ProGen` using these resources