ABodyBuilder2

Predict¶

Predict properties or scores for input sequences

from biolmai import BioLM
response = BioLM(
    entity="abodybuilder2",
    action="predict",
    params={},
    items=[
      {
        "H": "EVQLVESGGGLVQPGRSLRLSCAASGFTFDDYAMHWVRQAPGKGLEWVSAITWNSGHIDYADSVEGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCAKVSYLSTASSLDYWGQGTLVTVSS",
        "L": "DIQMTQSPSSLSASVGDRVTITCRASQDISNYLNWYQQKPGKAPKLLIYHTSRLHSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQGNTLPYTFGQGTKVEIK"
      },
      {
        "H": "QVQLQQSGPELEKPGASVKMSCKASGYTFTSYNMHWVKQTPGRGLEWIGAIYPGNGDTSYNQKFKGKATLTADKSSSTAYMQLSSLASEDSAVYYCARSNYYGSSYFDYWGQGTTLTVSS",
        "L": "QIVLSQSPAILSASPGEKVTMTCRASSSVNYMDWYQQKPGSSPKPWIYAPSNLASGVPVRFSGSGSGTSYSLTISRVEAEDAATYYCQQWSFNPPTFGAGTKLEIK"
      }
    ]
)
print(response)

curl -X POST https://biolm.ai/api/v3/abodybuilder2/predict/ \
  -H "Authorization: Token YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "items": [
    {
      "H": "EVQLVESGGGLVQPGRSLRLSCAASGFTFDDYAMHWVRQAPGKGLEWVSAITWNSGHIDYADSVEGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCAKVSYLSTASSLDYWGQGTLVTVSS",
      "L": "DIQMTQSPSSLSASVGDRVTITCRASQDISNYLNWYQQKPGKAPKLLIYHTSRLHSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQGNTLPYTFGQGTKVEIK"
    },
    {
      "H": "QVQLQQSGPELEKPGASVKMSCKASGYTFTSYNMHWVKQTPGRGLEWIGAIYPGNGDTSYNQKFKGKATLTADKSSSTAYMQLSSLASEDSAVYYCARSNYYGSSYFDYWGQGTTLTVSS",
      "L": "QIVLSQSPAILSASPGEKVTMTCRASSSVNYMDWYQQKPGSSPKPWIYAPSNLASGVPVRFSGSGSGTSYSLTISRVEAEDAATYYCQQWSFNPPTFGAGTKLEIK"
    }
  ]
}'

import requests

url = "https://biolm.ai/api/v3/abodybuilder2/predict/"
headers = {
    "Authorization": "Token YOUR_API_KEY",
    "Content-Type": "application/json"
}
payload = {
      "items": [
        {
          "H": "EVQLVESGGGLVQPGRSLRLSCAASGFTFDDYAMHWVRQAPGKGLEWVSAITWNSGHIDYADSVEGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCAKVSYLSTASSLDYWGQGTLVTVSS",
          "L": "DIQMTQSPSSLSASVGDRVTITCRASQDISNYLNWYQQKPGKAPKLLIYHTSRLHSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQGNTLPYTFGQGTKVEIK"
        },
        {
          "H": "QVQLQQSGPELEKPGASVKMSCKASGYTFTSYNMHWVKQTPGRGLEWIGAIYPGNGDTSYNQKFKGKATLTADKSSSTAYMQLSSLASEDSAVYYCARSNYYGSSYFDYWGQGTTLTVSS",
          "L": "QIVLSQSPAILSASPGEKVTMTCRASSSVNYMDWYQQKPGSSPKPWIYAPSNLASGVPVRFSGSGSGTSYSLTISRVEAEDAATYYCQQWSFNPPTFGAGTKLEIK"
        }
      ]
    }

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

library(httr)

url <- "https://biolm.ai/api/v3/abodybuilder2/predict/"
headers <- c("Authorization" = "Token YOUR_API_KEY", "Content-Type" = "application/json")
body <- list(
  items = list(
    list(
      H = "EVQLVESGGGLVQPGRSLRLSCAASGFTFDDYAMHWVRQAPGKGLEWVSAITWNSGHIDYADSVEGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCAKVSYLSTASSLDYWGQGTLVTVSS",
      L = "DIQMTQSPSSLSASVGDRVTITCRASQDISNYLNWYQQKPGKAPKLLIYHTSRLHSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQGNTLPYTFGQGTKVEIK"
    ),
    list(
      H = "QVQLQQSGPELEKPGASVKMSCKASGYTFTSYNMHWVKQTPGRGLEWIGAIYPGNGDTSYNQKFKGKATLTADKSSSTAYMQLSSLASEDSAVYYCARSNYYGSSYFDYWGQGTTLTVSS",
      L = "QIVLSQSPAILSASPGEKVTMTCRASSSVNYMDWYQQKPGSSPKPWIYAPSNLASGVPVRFSGSGSGTSYSLTISRVEAEDAATYYCQQWSFNPPTFGAGTKLEIK"
    )
  )
)

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

POST /api/v3/abodybuilder2/predict/¶

Predict endpoint for ABodyBuilder2.

Request Headers:

• Content-Type – application/json

• Authorization – Token YOUR_API_KEY

Request

• params (object, optional) — Configuration parameters:

  • plddt (bool, default: False) — Whether to include pLDDT scores in the response

  • seed (int, optional, default: 42) — Random seed for prediction consistency

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

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

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

Example request:

http Copy

POST /api/v3/abodybuilder2/predict/ HTTP/1.1
Host: biolm.ai
Authorization: Token YOUR_API_KEY
Content-Type: application/json

      {
  "items": [
    {
      "H": "EVQLVESGGGLVQPGRSLRLSCAASGFTFDDYAMHWVRQAPGKGLEWVSAITWNSGHIDYADSVEGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCAKVSYLSTASSLDYWGQGTLVTVSS",
      "L": "DIQMTQSPSSLSASVGDRVTITCRASQDISNYLNWYQQKPGKAPKLLIYHTSRLHSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQGNTLPYTFGQGTKVEIK"
    },
    {
      "H": "QVQLQQSGPELEKPGASVKMSCKASGYTFTSYNMHWVKQTPGRGLEWIGAIYPGNGDTSYNQKFKGKATLTADKSSSTAYMQLSSLASEDSAVYYCARSNYYGSSYFDYWGQGTTLTVSS",
      "L": "QIVLSQSPAILSASPGEKVTMTCRASSSVNYMDWYQQKPGSSPKPWIYAPSNLASGVPVRFSGSGSGTSYSLTISRVEAEDAATYYCQQWSFNPPTFGAGTKLEIK"
    }
  ]
}

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 format.

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

    • Outer array length: 2 (chains: heavy chain “H”, light chain “L”)

    • Inner array length: equal to the number of residues in the corresponding chain

    • Values range: 0.0–100.0 (higher values indicate higher local structure confidence)

Example response:

http Copy

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

      {
  "error": true,
  "status_code": 500,
  "message": "{\"error\":\"Uncaught exception: 'ImmuneBuilderPredictRequestItem' object has no attribute '_kind2'\",\"status_code\":500}"
}

Performance¶

• Batch Size and Sequence Length: - Supports a batch size of 1 - Maximum sequence length of 2048 amino acids

• GPU and Hardware Specifications: - Utilizes NVIDIA Tesla P100 GPUs for accelerated processing - Optimized for high throughput with GPU acceleration

• Performance and Speed: - ABodyBuilder2 offers rapid structure prediction, completing an antibody structure prediction in approximately 5 seconds on a Tesla P100 GPU - Significantly faster than AlphaFold-Multimer, which requires approximately 30 minutes on a GPU for a single structure - Designed for high-throughput applications, allowing for efficient processing of large datasets

• Accuracy and Predictive Performance: - ABodyBuilder2 demonstrates superior accuracy in predicting CDR-H3 loops with an RMSD of 2.81 Å, outperforming AlphaFold-Multimer’s 2.90 Å - Provides comparable or improved accuracy over other antibody-specific models like ABlooper, IgFold, and EquiFold - Ensures accurate side chain and chemical surface modeling, crucial for studying antigen binding

• Error Estimation and Reliability: - Generates an ensemble of four predictions per antibody, offering a confidence score for each residue - Facilitates filtering of inaccurately modeled regions by assessing prediction variability

• Comparative Model Performance: - ABodyBuilder2 is more accurate and faster than ESMFold and AlphaFold2 for antibody structure prediction - Outperforms general models like AlphaFold2 in antibody-specific tasks, while being more computationally efficient - Provides a balance of speed and accuracy, ideal for both research and commercial applications in biotechnology and pharmaceuticals

• Optimization and Scalability: - BioLM’s deployment of ABodyBuilder2 is optimized for scalability, ensuring consistent performance across varying workloads - Efficient resource utilization allows for cost-effective scaling in production environments

Applications¶

• Accurate prediction of antibody structures for therapeutic design, enabling the rapid development of new antibody drugs by providing detailed 3D models of antibody-antigen interactions.

• Structural modeling of T-cell receptors (TCRs) to enhance immunotherapy strategies, allowing researchers to design TCRs with improved specificity and binding affinity for cancer treatment.

• Nanobody structure prediction for biotechnological applications, offering insights into the unique binding properties of nanobodies, which are valuable for diagnostic tools and targeted drug delivery systems.

• High-throughput modeling of immune protein sequences from next-generation sequencing data, facilitating the exploration of vast immune repertoires and identification of promising candidates for vaccine development.

• Limitations include potential inaccuracies in predicting highly variable regions like CDR-H3 loops, which may require additional refinement for precise applications in drug design.

Limitations¶

• Maximum Sequence Length: Each antibody heavy (H) and light (L) chain sequence must not exceed 2048 amino acids.

• Batch Size: The API supports a maximum batch size of 1 sequence pair per request.

• ABodyBuilder2 is specifically optimized for antibody structure prediction, particularly the complementarity-determining region (CDR) loops. It is not suitable for general protein structure prediction or for other immune proteins such as nanobodies or T-cell receptors, for which NanoBodyBuilder2 or TCRBuilder2 should be used instead.

• While ABodyBuilder2 provides state-of-the-art accuracy for antibody CDR-H3 loops, predictions for unusually long or highly diverse CDR-H3 loops (e.g., longer than approximately 22 amino acids) may be less reliable.

• Although significantly faster than AlphaFold-Multimer, ABodyBuilder2 does not leverage evolutionary data or multiple sequence alignments, which may limit its accuracy in cases where such evolutionary context is critical.

• ABodyBuilder2 generates structural models but does not provide embeddings or encodings for downstream clustering or visualization tasks.

How We Use It¶

ABodyBuilder2 integrates into BioLM’s workflows by providing rapid and accurate predictions of antibody structures, significantly advancing protein design and engineering tasks. This model excels in predicting the complex CDR-H3 loops, a critical factor for antigen binding, and does so with enhanced speed compared to other methods. By enabling the accurate modeling of antibody structures, ABodyBuilder2 allows researchers to quickly iterate through design cycles, facilitating multi-round optimization and enhancing the precision of antibody engineering. This capability is crucial for accelerating research timelines and achieving practical outcomes in therapeutic development. Integration with other BioLM services, such as sequence embeddings and predictive models, further streamlines the protein engineering pipeline, providing a comprehensive suite of tools for scientific success.

• Accelerates research by reducing the time required for structural predictions, allowing for faster iteration and optimization.

• Enhances practical outcomes in therapeutic development through precise modeling of antibody structures.

Related¶

• NanoBodyBuilder2 – Predicts nanobody structures using similar deep-learning architecture, complementary for single-domain antibody modeling.

• TCRBuilder2 – Specialized deep-learning model for T-cell receptor structure prediction, closely related in architecture and immune protein focus.

• ImmuneFold Antibody – Alternative antibody structure prediction method, useful for comparative modeling and validation of ABodyBuilder2 predictions.

• ABodyBuilder3 pLDDT – Next-generation antibody structure prediction model providing confidence estimates, complementary for assessing model reliability.

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.