Evaluating Suitable Embeddings for Benchmarking Biological Foundation Models state or esm?

[shiyegao]

Hello,

I’ve been working on benchmarking biological foundation models, and I’m exploring which embeddings are most suitable for my study. I noticed that the state embedding has been used for lddt prediction in Line.

Would it be appropriate to use the state embedding for my research on representations? Or would the msa embedding, which is applied to masked amino acids prediction in Line, be a better option?

Thank you for your time and insights. I’m looking forward to your thoughts!