ESM-1v is a 650M-parameter protein language model for zero-shot prediction of mutational effects on protein function, scoring substitutions directly from single amino acid sequences without task-specific training or MSAs. The API accepts batches of 1–5 sequences (length ≤ 512, optionally containing a single <mask> token) and returns per-position log-odds scores over the 20 canonical amino acids from one selected ESM-1v model (n1–n5) or a GPU-backed ensemble (all). Typical uses include variant prioritization, protein engineering, and enzyme or antibody optimization.

Predict

Predict properties or scores for input sequences

python 
from biolmai import BioLM
response = BioLM(
    entity="esm1v-all",
    action="predict",
    params={},
    items=[
      {
        "sequence": "ACDG<mask>HIKLMN"
      },
      {
        "sequence": "XPQRS<mask>FGT"
      }
    ]
)
print(response)
bash 
curl -X POST https://biolm.ai/api/v3/esm1v-all/predict/ \
  -H "Authorization: Token YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "params": {},
  "items": [
    {
      "sequence": "ACDG<mask>HIKLMN"
    },
    {
      "sequence": "XPQRS<mask>FGT"
    }
  ]
}'
python 
import requests

url = "https://biolm.ai/api/v3/esm1v-all/predict/"
headers = {
    "Authorization": "Token YOUR_API_KEY",
    "Content-Type": "application/json"
}
payload = {
      "params": {},
      "items": [
        {
          "sequence": "ACDG<mask>HIKLMN"
        },
        {
          "sequence": "XPQRS<mask>FGT"
        }
      ]
    }

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

url <- "https://biolm.ai/api/v3/esm1v-all/predict/"
headers <- c("Authorization" = "Token YOUR_API_KEY", "Content-Type" = "application/json")
body <- list(
  params = list(),
  items = list(
    list(
      sequence = "ACDG<mask>HIKLMN"
    ),
    list(
      sequence = "XPQRS<mask>FGT"
    )
  )
)

res <- POST(url, add_headers(.headers = headers), body = body, encode = "json")
print(content(res))
POST /api/v3/esm1v-all/predict/

Predict endpoint for ESM-1v.

Request Headers:

Request

  • params (object, optional) — Configuration parameters:

    • model_number (string, enum: [“n1”, “n2”, “n3”, “n4”, “n5”, “all”], default: “all”) — ESM-1v model variant used to score each input sequence

  • items (array of objects, min: 1, max: 5) — Input sequences:

    • sequence (string, min length: 1, max length: 512, required) — Protein sequence containing exactly one “<mask>” token, using extended amino acid characters plus “<mask>”

Example request:

http 
POST /api/v3/esm1v-all/predict/ HTTP/1.1
Host: biolm.ai
Authorization: Token YOUR_API_KEY
Content-Type: application/json

      {
  "params": {},
  "items": [
    {
      "sequence": "ACDG<mask>HIKLMN"
    },
    {
      "sequence": "XPQRS<mask>FGT"
    }
  ]
}
Status Codes:

Response

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

    • esm1v-n1 (array of objects) — Predictions from ESM-1v model variant n1

      • token (int) — Integer token index for the amino acid at the masked position

      • token_str (string) — Single-letter amino acid code or special token corresponding to token

      • score (float, range: 0.0-1.0) — Predicted probability for token at the masked position

      • sequence (string) — Input sequence with <mask> replaced by token_str

    • esm1v-n2 (array of objects) — Predictions from ESM-1v model variant n2

      • token (int) — Integer token index for the amino acid at the masked position

      • token_str (string) — Single-letter amino acid code or special token corresponding to token

      • score (float, range: 0.0-1.0) — Predicted probability for token at the masked position

      • sequence (string) — Input sequence with <mask> replaced by token_str

    • esm1v-n3 (array of objects) — Predictions from ESM-1v model variant n3

      • token (int) — Integer token index for the amino acid at the masked position

      • token_str (string) — Single-letter amino acid code or special token corresponding to token

      • score (float, range: 0.0-1.0) — Predicted probability for token at the masked position

      • sequence (string) — Input sequence with <mask> replaced by token_str

    • esm1v-n4 (array of objects) — Predictions from ESM-1v model variant n4

      • token (int) — Integer token index for the amino acid at the masked position

      • token_str (string) — Single-letter amino acid code or special token corresponding to token

      • score (float, range: 0.0-1.0) — Predicted probability for token at the masked position

      • sequence (string) — Input sequence with <mask> replaced by token_str

    • esm1v-n5 (array of objects) — Predictions from ESM-1v model variant n5

      • token (int) — Integer token index for the amino acid at the masked position

      • token_str (string) — Single-letter amino acid code or special token corresponding to token

      • score (float, range: 0.0-1.0) — Predicted probability for token at the masked position

      • sequence (string) — Input sequence with <mask> replaced by token_str

Example response:

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

      {
  "results": [
    {
      "esm1v-n1": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.09930282086133957,
          "sequence": "A C D G L H I K L M N"
        },
        {
          "token": 12,
          "token_str": "I",
          "score": 0.07655816525220871,
          "sequence": "A C D G I H I K L M N"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n2": [
        {
          "token": 8,
          "token_str": "S",
          "score": 0.0869770348072052,
          "sequence": "A C D G S H I K L M N"
        },
        {
          "token": 4,
          "token_str": "L",
          "score": 0.08455677330493927,
          "sequence": "A C D G L H I K L M N"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n4": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.09174387156963348,
          "sequence": "A C D G L H I K L M N"
        },
        {
          "token": 8,
          "token_str": "S",
          "score": 0.07290017604827881,
          "sequence": "A C D G S H I K L M N"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n5": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.09874340891838074,
          "sequence": "A C D G L H I K L M N"
        },
        {
          "token": 8,
          "token_str": "S",
          "score": 0.07558094710111618,
          "sequence": "A C D G S H I K L M N"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n3": [
        {
          "token": 6,
          "token_str": "G",
          "score": 0.10201912373304367,
          "sequence": "A C D G G H I K L M N"
        },
        {
          "token": 4,
          "token_str": "L",
          "score": 0.08753731101751328,
          "sequence": "A C D G L H I K L M N"
        },
        "... (truncated for documentation)"
      ]
    },
    {
      "esm1v-n1": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.10145029425621033,
          "sequence": "X P Q R S L F G T"
        },
        {
          "token": 8,
          "token_str": "S",
          "score": 0.09379883855581284,
          "sequence": "X P Q R S S F G T"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n2": [
        {
          "token": 8,
          "token_str": "S",
          "score": 0.0989096611738205,
          "sequence": "X P Q R S S F G T"
        },
        {
          "token": 4,
          "token_str": "L",
          "score": 0.09306105226278305,
          "sequence": "X P Q R S L F G T"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n4": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.3064996600151062,
          "sequence": "X P Q R S L F G T"
        },
        {
          "token": 12,
          "token_str": "I",
          "score": 0.11773010343313217,
          "sequence": "X P Q R S I F G T"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n5": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.6074760556221008,
          "sequence": "X P Q R S L F G T"
        },
        {
          "token": 18,
          "token_str": "F",
          "score": 0.05280425772070885,
          "sequence": "X P Q R S F F G T"
        },
        "... (truncated for documentation)"
      ],
      "esm1v-n3": [
        {
          "token": 4,
          "token_str": "L",
          "score": 0.10762723535299301,
          "sequence": "X P Q R S L F G T"
        },
        {
          "token": 6,
          "token_str": "G",
          "score": 0.10102758556604385,
          "sequence": "X P Q R S G F G T"
        },
        "... (truncated for documentation)"
      ]
    }
  ]
}

Performance

  • GPU-accelerated inference for the 5-model ESM-1v ensemble (esm1v-all) runs on NVIDIA T4 GPUs, while individual ensemble members (esm1v-n1–n5) are served on lightweight CPU instances optimized for high-throughput scoring.

  • Zero-shot mutation effect prediction achieves average |Spearman ρ| ≈ 0.51 across 41 deep mutational scanning datasets, matching state-of-the-art unsupervised MSA-based methods (DeepSequence and EVMutation, each ≈ 0.51) without per-protein training or MSA generation.

  • Compared to earlier single-sequence models available on BioLM (ESM-1b |ρ| ≈ 0.46, ProtBERT |ρ| ≈ 0.43 on the same benchmarks), ESM-1v provides higher zero-shot accuracy; the 5-model ensemble attains the strongest correlations and exceeds DeepSequence on 17/41 datasets.

  • Inference is substantially more efficient than MSA-dependent architectures (e.g., MSA Transformer, DeepSequence, EVMutation): ESM-1v uses masked forward passes on single, unaligned sequences, enabling faster, lower-cost scoring for large mutational libraries while approaching or matching their predictive performance.

Applications

  • Zero-shot scoring of single or multiple amino acid substitutions to prioritize variants in protein engineering campaigns, enabling rapid in silico filtering before deep mutational scanning or focused library construction; most reliable when variants remain close to natural homologs in length and composition, and less reliable for highly unnatural sequences or poorly constrained regions.

  • Computational triage of industrial enzyme libraries to enrich for likely functional or stabilizing mutations before expression and assay, reducing experimental load for teams optimizing catalytic efficiency, specificity, or process robustness; performance may degrade for proteins with very limited representation in natural sequence databases or atypical cofactors.

  • Identifying functionally constrained or putative active-site residues by ranking per-position mutation effects, guiding where to focus saturation mutagenesis or combinatorial design in workflows for metabolic pathway enzymes, transporters, or receptors; less informative for positions with weak evolutionary constraints, long-range epistasis, or strong context-dependent allosteric effects.

  • Variant effect scoring to support assessment of protein-coding changes in research, diagnostics development, or safety workflows, providing an additional sequence-based line of evidence for classifying missense variants as more likely benign vs. more likely damaging; should be combined with structural, population, and clinical evidence and not used as a sole decision criterion.

  • Pre-screening development or manufacturing variants (e.g., manufacturability edits or sequence optimizations) to avoid substitutions predicted to strongly reduce protein fitness, helping biopharma teams de-risk sequence changes that might impact yield or stability; not optimal for predicting effects driven primarily by post-translational modifications, formulation conditions, or non-sequence process parameters.

Limitations

  • Maximum Sequence Length: Input protein sequences in the sequence field must be <= 512 amino acids (including the single <mask> token). Longer proteins must be truncated or split into overlapping windows before calling the predictor endpoint.

  • Batch Size: Each predictor request accepts at most 5 items in the items list. Larger mutational scans must be split across multiple requests and merged client-side.

  • GPU Type: GPU acceleration (gpu=ModalGPU.T4) is available only when model is set to the ensemble all. Individual models n1–n5 use gpu=None and run on CPU only, which may be slower for large libraries.

  • Scoring Task Only: The API returns per-amino-acid log-probability scores around the single <mask> site as ESM1vPredictResponseLabel entries (token, token_str, score, sequence). It does not expose structure prediction, sequence generation, or embeddings.

  • Single Masked Site: Each sequence may contain at most one <mask> token (enforced by SingleOccurrenceOf). Scoring mutations at multiple positions requires separate request items or calls, one masked position per sequence.

  • Biological Domain and Training Data: ESM-1v is optimized for zero-shot variant effect prediction on natural-like protein sequences. Scores on highly synthetic, non-protein, or very low-homology sequences can be less reliable and should be interpreted together with experimental or structural data.

How We Use It

BioLM uses ESM-1v as a zero-shot variant-effect scorer in protein engineering workflows to prioritize large mutational libraries before lab work. Standardized, API-based access to per-position mutation scores enables rapid ranking of single and simple combinatorial variants, and these scores are integrated with embedding-based clustering, supervised fitness models, and structure- and developability-aware analyses to guide iterative design cycles for antibodies, enzymes, and other proteins.

  • Enables rapid triage of candidate protein sequences prior to synthesis and experimental screening

  • Integrates with embedding-based and 3D-structure workflows to align sequence-level ranking with downstream functional and biophysical objectives

References

  • Meier, J., Rao, R., Verkuil, R., Liu, J., Sercu, T., & Rives, A. (2021). Language models enable zero-shot prediction of the effects of mutations on protein function. bioRxiv, 2021.07.09.450648. https://doi.org/10.1101/2021.07.09.450648