Inquiry About Your Research and Published Code #10
Description
I am deeply honored to have read your paper "A general temperature-guided language model to design proteins of enhanced stability and activity." The content is profound and highly inspiring, and I have learned a great deal from it.
While applying your research findings in practice, I have encountered some questions and sincerely hope to receive your guidance.
In your published code, I noticed that you provided a trained model (AI4Protein/Prime_690M) which shows excellent performance in OGT prediction tasks. However, my goal is to optimize protein activity, which requires combining the MLM module mentioned in your paper with the OGT module that has been fine-tuned multiple times through wet lab experiments. Regarding this, I have several questions:
In your published "AI4Protein/Prime_690M" model, has the "Transformer encoder" section been fine-tuned with OGT data, making it more inclined to predict OGT from protein sequences while weakening the model's MLM prediction capabilities?
I need to optimize protein mutations at multiple sites and need to use the MLM and OGT modules mentioned in the paper. Are there any relevant code or resources I can refer to?
If possible, could you share the PLM training process code for PRIME? This would help better understand and reproduce your research.
Thank you so much and look forward to your reply!