ABodyBuilder3 pLDDT

Predict¶

Predict 3D structure and pLDDT scores for an antibody’s heavy (H) and light (L) chain.

from biolmai import BioLM
response = BioLM(
    entity="abodybuilder3-plddt",
    action="predict",
    params={
      "plddt": true,
      "seed": 123
    },
    items=[
      {
        "H": "EVQLVESGGGLVKPGGSLR",
        "L": "DIQMTQSPASLSASVGDR"
      }
    ]
)
print(response)

curl -X POST https://biolm.ai/api/v3/abodybuilder3-plddt/predict/ \
  -H "Authorization: Token YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "params": {
    "plddt": true,
    "seed": 123
  },
  "items": [
    {
      "H": "EVQLVESGGGLVKPGGSLR",
      "L": "DIQMTQSPASLSASVGDR"
    }
  ]
}'

import requests

url = "https://biolm.ai/api/v3/abodybuilder3-plddt/predict/"
headers = {
    "Authorization": "Token YOUR_API_KEY",
    "Content-Type": "application/json"
}
payload = {
      "params": {
        "plddt": true,
        "seed": 123
      },
      "items": [
        {
          "H": "EVQLVESGGGLVKPGGSLR",
          "L": "DIQMTQSPASLSASVGDR"
        }
      ]
    }

response = requests.post(url, headers=headers, json=payload)
print(response.json())

library(httr)

url <- "https://biolm.ai/api/v3/abodybuilder3-plddt/predict/"
headers <- c("Authorization" = "Token YOUR_API_KEY", "Content-Type" = "application/json")
body <- list(
  params = list(
    plddt = TRUE,
    seed = 123
  ),
  items = list(
    list(
      H = "EVQLVESGGGLVKPGGSLR",
      L = "DIQMTQSPASLSASVGDR"
    )
  )
)

res <- POST(url, add_headers(.headers = headers), body = body, encode = "json")
print(content(res))

POST /api/v3/abodybuilder3-plddt/predict/¶

Predict endpoint for ABodyBuilder3 pLDDT.

Request Headers:

• Content-Type – application/json

• Authorization – Token YOUR_API_KEY

Request

• params (object, optional) — Configuration parameters:

  • plddt (bool, default: False) — Flag indicating whether to compute per-residue predicted Local Distance Difference Test (pLDDT) confidence scores.

  • seed (int, optional, default: 42) — Random seed for reproducibility.

• items (array of objects, min: 1, max: 1) — Input antibody sequences:

  • H (string, required, min length: 1, max length: 2048) — Heavy chain amino acid sequence using extended amino acid alphabet.

  • L (string, required, min length: 1, max length: 2048) — Light chain amino acid sequence using extended amino acid alphabet.

Example request:

http Copy

POST /api/v3/abodybuilder3-plddt/predict/ HTTP/1.1
Host: biolm.ai
Authorization: Token YOUR_API_KEY
Content-Type: application/json

      {
  "params": {
    "plddt": true,
    "seed": 123
  },
  "items": [
    {
      "H": "EVQLVESGGGLVKPGGSLR",
      "L": "DIQMTQSPASLSASVGDR"
    }
  ]
}

Status Codes:

• 200 OK – Successful response

• 400 Bad Request – Invalid input

• 500 Internal Server Error – Internal server error

Response

• results (array of objects) — One result per input item, in the order requested:

  • pdb (string) — Predicted antibody structure in standard PDB file format.

  • plddt (array of arrays of floats, optional) — Predicted Local Distance Difference Test (pLDDT) confidence scores per residue.

    • Outer array dimension: 2 (one array per antibody chain: heavy [H] and light [L]).

    • Inner array dimension: equal to the length of each input sequence (H and L chains).

    • Float values range: 0.0–100.0, higher scores indicate higher confidence.

Example response:

http Copy

HTTP/1.1 200 OK
Content-Type: application/json

      {
  "results": [
    {
      "pdb": "REMARK   1 CREATED WITH OPENMM 7.7, 2025-06-17\nATOM      1  N   GLU H   0     -12.538   1.205   0.765  1.00  0.00           N  \nATOM      2  H   GLU H   0     -13.550   0.585   0.591  1.00  0.00      ... (truncated for documentation)",
      "plddt": [
        [
          -2.409224510192871,
          -3.094425916671753,
          "... (truncated for documentation)"
        ],
        [
          -3.281006336212158,
          -4.06817626953125,
          "... (truncated for documentation)"
        ],
        "... (truncated for documentation)"
      ]
    }
  ]
}

Performance¶

• ABodyBuilder3 pLDDT runs exclusively on CPU resources (2 vCPUs, 8GB RAM); no GPU acceleration is required or utilized.

• ABodyBuilder3 pLDDT provides residue-level confidence estimates (pLDDT scores) for antibody structure predictions generated by ABodyBuilder3 Language, enabling accurate assessment of model quality and reliability.

• ABodyBuilder3 pLDDT achieves comparable accuracy in confidence estimation to AlphaFold2’s pLDDT metric, but with significantly reduced computational requirements, making it suitable for high-throughput antibody modeling workflows.

• Compared to AlphaFold2, which requires GPU resources and extensive sequence database searches, ABodyBuilder3 pLDDT is optimized for rapid inference on CPU-only hardware, delivering residue-level confidence scores in seconds per antibody structure.

• ABodyBuilder3 pLDDT predictions are returned as per-residue floating-point values (List[List[float]]), corresponding directly to residue positions in the input antibody sequences (heavy and light chains).

• Input sequences must be valid amino acid sequences provided as separate heavy (“H”) and light (“L”) chain strings.

• Output consists solely of per-residue pLDDT confidence scores aligned to the provided antibody sequences; no structural coordinates (PDB files) are returned by this endpoint.

Applications¶

• Rapid antibody candidate screening and prioritization by predicting structural confidence (pLDDT), enabling biotech companies to efficiently filter large antibody libraries and focus experimental resources on candidates with high predicted structural reliability; not optimal for final therapeutic validation without experimental confirmation.

• Antibody affinity maturation workflows, where structural confidence scores (pLDDT) help protein engineers identify stable backbone conformations, guiding targeted mutagenesis to improve antigen binding; particularly useful for companies developing antibody therapeutics requiring precise binding site optimization.

• Computational epitope mapping and characterization by assessing antibody structural reliability (pLDDT), allowing researchers to confidently identify structurally stable CDR loops for antigen interaction studies; beneficial for vaccine developers and antibody-based diagnostic companies aiming to understand antigen-antibody interfaces.

• Antibody humanization projects, using pLDDT confidence scores to evaluate structural integrity of humanized antibody variants, thus accelerating the selection of stable, human-compatible therapeutic antibodies; less suitable for predicting immunogenicity or functional activity without additional assays.

• Structural quality control in antibody database curation, employing pLDDT scores to automatically flag low-confidence structural predictions, ensuring that biotech companies maintain reliable antibody structure repositories for downstream modeling and computational analysis; not intended to replace manual expert review entirely.

Limitations¶

• Maximum Sequence Length: The API accepts antibody heavy (“H”) and light (“L”) chain sequences up to 2048 amino acids each. Longer sequences cannot be processed and must be truncated or split prior to submission.

• Batch Size: The API supports a fixed batch_size of exactly 1 item per request. For larger datasets, requests must be submitted sequentially.

• Model Type (plddt): The plddt option returns per-residue confidence scores reflecting prediction uncertainty. However, low predicted error scores do not guarantee structural accuracy; they are best used to filter out high-error predictions rather than confirming correctness.

• Structural Accuracy: While ABodyBuilder3 achieves high accuracy for antibody frameworks and most CDR loops, the CDR-H3 loop remains challenging to predict reliably. Predictions for CDR-H3 loops may have higher RMSD (average ~2.81 Å), which should be considered when interpreting results.

• Molecule Type: ABodyBuilder3 is specifically optimized for antibody variable domain structures. It is not suitable for general protein structure prediction, nanobodies, or T-cell receptors, for which specialized models (e.g., NanoBodyBuilder2, TCRBuilder2) are recommended.

• Ranking and Filtering: Due to computational cost and specialized antibody focus, ABodyBuilder3 is not optimal for initial high-throughput filtering or ranking of large sequence libraries. Faster, less computationally intensive models (e.g., ESMFold) are recommended for early-stage pipeline steps involving millions of sequences.

How We Use It¶

ABodyBuilder3 pLDDT enables rapid, accurate assessment of antibody structural confidence directly within BioLM’s antibody design and optimization workflows, providing residue-level quality metrics to prioritize candidates for synthesis and experimental validation. Integrated seamlessly into pipelines alongside predictive and generative models, sequence embeddings, and biophysical property estimators, this algorithm accelerates antibody maturation cycles by efficiently filtering sequences with suboptimal structural predictions.

• Facilitates data-driven selection of antibody candidates, saving experimental resources.

• Integrates smoothly with BioLM’s predictive modeling and structure-based optimization tools, streamlining antibody engineering projects.

Related¶

• NanoBodyBuilder2 – Predicts nanobody structures with similar accuracy and speed, complementing ABodyBuilder3 pLDDT for single-domain antibodies.

• TCRBuilder2 – Provides accurate structure predictions for T-cell receptors, complementing antibody predictions from ABodyBuilder3 pLDDT.

• ImmuneFold Antibody – Offers alternative antibody structure predictions, useful for cross-validation with ABodyBuilder3 pLDDT results.

• ABodyBuilder3 Language – Generates antibody structures using language-model-based approaches, complementary to the pLDDT confidence metric predictions.

References¶

• Abanades, B., Wong, W. K., Boyles, F., Georges, G., Bujotzek, A., & Deane, C. M. (2023). ImmuneBuilder: Deep-Learning models for predicting the structures of immune proteins. Communications Biology.