Paper: Large language models generate functional protein sequences across diverse families
I am currently working towards this paper. Check out my learning progress here: https://twitter.com/xariusrke/status/1621403313651728386
- [DONE] fast2dict
- Extract multiple properties of a protein sequence
- Tokenize the control tag
- Generate training data from dataset
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
I implemented ProGen using these resources