Great Work and Better Models
agemagician opened this issue · 3 comments
Hi @shamika and @mani-aiml,
It's really great to see our models are integrated into amazon segamaker pipeline 😄
We have released 2 new models which provides even better results than "ProtBert":
The models are "ProtBert-BFD" and "ProtT5-XL-BFD":
https://github.com/agemagician/ProtTrans/blob/master/Embedding/PyTorch/Advanced/ProtBert-BFD.ipynb
https://github.com/agemagician/ProtTrans/blob/master/Embedding/PyTorch/Advanced/ProtT5-XL-BFD.ipynb
They were trained on much bigger datasets. "ProtBert-BFD" has the same number of parameters as "ProtBert", but "ProtT5-XL-BFD" has 1.5B parameter for it's encoder part.
Let us know if you need any clarification or help.
I would also recommend to have some examples for amino-acid level classification like secondary structure protein prediction.
You can find an example here:
https://github.com/agemagician/ProtTrans/blob/master/Fine-Tuning/ProtBert-BFD-FineTune-SS3.ipynb
Hi @shamika and @mani-aiml,
It's really great to see our models are integrated into amazon segamaker pipeline 😄
We have released 2 new models which provides even better results than "ProtBert":
The models are "ProtBert-BFD" and "ProtT5-XL-BFD":
https://github.com/agemagician/ProtTrans/blob/master/Embedding/PyTorch/Advanced/ProtBert-BFD.ipynb
https://github.com/agemagician/ProtTrans/blob/master/Embedding/PyTorch/Advanced/ProtT5-XL-BFD.ipynbThey were trained on much bigger datasets. "ProtBert-BFD" has the same number of parameters as "ProtBert", but "ProtT5-XL-BFD" has 1.5B parameter for it's encoder part.
Let us know if you need any clarification or help.
@agemagician : Thank you Ahmed for your input. We are using Rostlab/prot_bert_bfd_localization for it. I think it's a good idea to try out other models as mentioned by you as well.
I would also recommend to have some examples for amino-acid level classification like secondary structure protein prediction.
You can find an example here:
https://github.com/agemagician/ProtTrans/blob/master/Fine-Tuning/ProtBert-BFD-FineTune-SS3.ipynb
@agemagician Thank you Ahmed for sharing it, I will look into the amino-acid classification to see, if we can add that too.