Use MIF-ST

The input protein structure to predict the amino acid sequence of. Acceptable input file formats include pdb only.

The number of output sequences to generate.

The percentage of amino acids to mask from the model. We typically recommend between 0.3 to 0.75. Lower values means less amino acids will be masked from the model resulting in output sequences that are more similar to the input structure. MIF-ST typically struggles when values are less than 0.3 as the model is trained to be heavily dependent on sequence information.

The sampling temperature to use (recommended is 0.9). Lower values will result in less randomness and more conserved sequences, higher values will result in greater diversity and randomness. For more details visit https://docs.cohere.com/docs/temperature

The top k number of amino acids to sample from per residue (recommended is 4). Greater values will result in more diverse sequences, while lower values will result in more conservative estimates. Setting this to 0 will disable top k sampling.

Nucleus (top-p) dynamic sampling method (recommended is 0.5). Uses the top amino acids who's sum of likelihoods do not exceed this value. Greater values will result in more diverse sequences, while lower values will result in more conservative estimates. Setting this to 0 will disable top p sampling. For more details visit https://docs.cohere.com/docs/controlling-generation-with-top-k-top-p