Generate new sequence by fixing the length

[shrimonmuke0202]

Hi,

Your work is fantastic!

I have a question: I want to generate a sequence of proteins by providing the length as input.

For example:
Generate a sequence of length 100.

How can I do this?

Thanks and regards,
Shrimon Mukherjee

[Lyu6PosHao]

Hi,
Thanks for your recognization!
For decoder-only autoregressive models such as ProLLaMA, there is no direct way to control the output length to fixed values such as 100.
One possible approach is to let the model generate many sequences, and then manually filter out those with a length of 100.

Kind regards

[shrimonmuke0202]

Thanks for your answer. I want to produce the results present in the paper for my work, in particular the Table 2. could you share the code used to calculate the metrics like Tm score, RMSD score

[Lyu6PosHao]

Please refer to https://github.com/steineggerlab/foldseek. You could easily use the tools provided by foldseek to calculate TMscore, RMSD, etc.

[shrimonmuke0202]

Thanks, how you perform the evaluation process i.e; how many new sequences are generated to performa the evaluation process?

[shrimonmuke0202]

Also could you share the test set. It will help for me to compare between ProLlama and our proposed model.

[Lyu6PosHao]

1000 protein sequences used for each model in Table 2. We use models in Table2 to generate 1000 proteins respectively.

You could use ProLLaMA_Stage_1 to generate 1000 sequecens unconditionally throught scripts/infer.py. And then compare the proteins generated by ProLLaMA with your proposed model.

Best Regards