Pool PaRTI-generated amino acid level normalized importance scores (weights for the weighted averaging) for ESM-2 650M and ProtBERT

Please see the expanded version (with sequence embeddings) in zenodo.org/records/15036725

We generated residue (or token) level averaging weights or importance scores for each human protein sequence on UniProt. We used two different protein language models whose outcomes are highly correlated with one another. The protein sequence importance arrays are indexed by their UniProt accessions in the dictionaries saved as npz files. In other words, the protein UniProt accessions are the keys, and the corresponding arrays whose lengths are equal to the sequence lengths are the values.

If you would like to inquire about a protein sequence not found in these dictionaries, please reach out to us.