Protein Features and Structure¶

See the FAQ for dependencies, checkpoints and runtime configuration.

1) Label biological categories of protein SAE features¶

Purpose: Label biological categories of protein SAE features.
Method: Use ESM to run the selected trained model on the supplied input.
API： dnakit.databases.esm_describe_sae_feature(*, feature_id, sae_model=None, model=None, n_proteins=None, top_residues_per_protein=None, use_cache=None, verbose=True)
Input: Provide feature_id. Configure: ESM_API_KEY.
Output: Returns a mapping with feature_id, sae_model, category, confidence and related fields.
Example code:

import dnakit

result = dnakit.databases.esm_describe_sae_feature(
    feature_id=6404,
    n_proteins=5,
    verbose=False,
)
print(result)

Example output: The response has the following shape; records depend on the input and source.

Response-shape excerpt; <…> denotes type placeholders, not a captured query result.

{
  "feature_id": "<integer>",
  "sae_model": "<string>",
  "category": "<string>",
  "confidence": "<number>",
  "n_proteins_with_activation": "<integer>",
  "n_proteins_analyzed": "<integer>",
  "category_vote_counts": "<object>",
  "supporting_evidence": [
    {
      "protein": "<string>",
      "position_1based": "<integer>",
      "activation": "<number>",
      "uniprot_types": "<array>",
      "uniprot_categories": "<array>"
    }
  ]
}

Full parameters and usage notes

2) Aggregate protein SAE features over a residue interval¶

Purpose: Aggregate protein SAE features over a residue interval.
Method: Use ESM to run the selected trained model on the supplied input.
API： dnakit.databases.esm_get_region_sae_features(*, sequence, start_position, end_position, top_k_features=None, model=None, sae_model=None, verbose=True)
Input: Provide sequence, start_position, end_position. Configure: ESM_API_KEY.
Output: Returns a mapping with sequence_length, region, region_length, n_features_active_in_region and related fields.
Example code:

import dnakit

example_sequence = 'MTEYKLVVVGAGGVGKSALTIQLIQNHFVDEYDPTIEDSYRKQVVIDGETCLLDILDTAGQEEYSAMRDQYMRTGEGFLCVFAINNTKSFEDIHHYREQIKRVKDSEDVPMVLVGNKCDLPSRTVDTKQAQDLARSYGIPFIETSAKTRQGVDDAFYTLVREIRKHKEK'

result = dnakit.databases.esm_get_region_sae_features(
    sequence=example_sequence,
    start_position=10,
    end_position=17,
    top_k_features=10,
    verbose=False,
)
print(result)

Example output: The response has the following shape; records depend on the input and source.

Response-shape excerpt; <…> denotes type placeholders, not a captured query result.

{
  "sequence_length": "<integer>",
  "region": [
    "<value>"
  ],
  "region_length": "<integer>",
  "n_features_active_in_region": "<integer>",
  "top_features": [
    "<value>"
  ]
}

Full parameters and usage notes

3) Extract per-residue protein SAE activations¶

Purpose: Extract per-residue protein SAE activations.
Method: Use ESM to run the selected trained model on the supplied input.
API： dnakit.databases.esm_get_sae_features(*, sequence, model=None, sae_model=None, position=None, window=None, top_k_per_residue=None, verbose=True)
Input: Provide sequence. Configure: ESM_API_KEY.
Output: Returns a mapping with sequence_length, model, sae_model, residues_returned and related fields.
Example code:

import dnakit

example_sequence = 'MEEPQSDPSVEPPLSQETFSDLWKLLPENNVLSPLPSQAMDDLMLSPDDIEQWFTEDPGPDEAPRMPEAAPPVAPAPAAPTPAAPAPAPSWPLSSSVPSQKTYQGSYGFRLGFLHSGTAKSVTCTYSPALNKMFCQLAKTCPVQLWVDSTPPPGTRVRAMAIYKQSQHMTEVVRRCPHHERCSDSDGLAPPQHLIRVEGNLRVEYLDDR'

result = dnakit.databases.esm_get_sae_features(
    sequence=example_sequence,
    position=175,
    window=8,
    top_k_per_residue=16,
    verbose=False,
)
print(result)

Example output: The response has the following shape; records depend on the input and source.

Response-shape excerpt; <…> denotes type placeholders, not a captured query result.

{
  "sequence_length": "<integer>",
  "model": "<string>",
  "sae_model": "<string>",
  "residues_returned": "<integer>",
  "position": "<integer>",
  "window": "<integer>",
  "top_k_per_residue": "<integer>",
  "activations": [
    {
      "residue_idx_1based": "<integer>",
      "active_features": "<array>"
    }
  ]
}

Full parameters and usage notes

4) Score protein sequence log-probabilities with ESMC¶

Purpose: Score protein sequence log-probabilities with ESMC.
Method: Use ESM to run the selected trained model on the supplied input.
API： dnakit.databases.esm_score_sequence(*, sequence, model=None, verbose=True)
Input: Provide sequence. Configure: ESM_API_KEY.
Output: Returns a mapping with status, model, sequence, sequence_length and related fields.
Example code:

import dnakit

result = dnakit.databases.esm_score_sequence(
    sequence='MKTAYIAKQRQISFVKSHFSRQ',
    model='esmc-300m-2024-12',
    verbose=False,
)
print(result)

Example output: The response has the following shape; records depend on the input and source.

Response-shape excerpt; <…> denotes type placeholders, not a captured query result.

{
  "status": "<string>",
  "model": "<string>",
  "sequence": "<string>",
  "sequence_length": "<integer>",
  "mean_log_probability": "<number>",
  "per_residue_log_probabilities": [
    "<number>"
  ]
}

Full parameters and usage notes

5) Predict protein structures with AlphaFold2¶

Purpose: Predict protein structures with AlphaFold2.
Method: Use NvidiaNIM to query or submit the supplied input and parse the response.
API： dnakit.databases.nvidia_nim_alphafold2(*, sequence, algorithm=None, databases=None, e_value=None, iterations=None, relax_prediction=None, skip_template_search=None, verbose=True)
Input: Provide sequence. Configure: NVIDIA_API_KEY.
Output: Returns predicted PDB coordinates and service-provided confidence and model metrics.
Example code:

import dnakit

result = dnakit.databases.nvidia_nim_alphafold2(
    sequence='MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH',
    verbose=False,
)
print(result)

Example output: Returns predicted PDB coordinates and service-provided confidence and model metrics.

Full parameters and usage notes

6) Predict protein structures from sequence with ESMFold¶

Purpose: Predict protein structures from sequence with ESMFold.
Method: Use NvidiaNIM to query or submit the supplied input and parse the response.
API： dnakit.databases.nvidia_nim_esmfold(*, sequence, verbose=True)
Input: Provide sequence. Configure: NVIDIA_API_KEY.
Output: Returns a string.
Example code:

import dnakit

result = dnakit.databases.nvidia_nim_esmfold(
    sequence='ARNDCQEGHILKMFPSTWYV',
    verbose=False,
)
print(result)

Example output: The response has the following shape; records depend on the input and source.

Response-shape excerpt; <…> denotes type placeholders, not a captured query result.

"<string>"

Full parameters and usage notes