ProtGPT3-MSA is a multiple-sequence, homolog-conditioned autoregressive protein language model. It is part of the ProtGPT3 family, an open-source suite of promptable and aligned protein language models for protein sequence generation.

Unlike the single-sequence ProtGPT3 models, ProtGPT3-MSA can be prompted with sets of homologous protein sequences, enabling few-shot, family-conditioned protein generation without task-specific fine-tuning. At inference, users can provide homologous protein sequences as context and generate additional family-consistent sequences.

ProtGPT3-MSA was trained to autoregressively predict sets of 16 concatenated protein sequences, separated by a special token <s> (i.e., marking the protein boundaries). Therefore, at inference, the model should be prompted with at most 15 concatenated protein sequences.

For more details on how to use ProtGPT3-MSA check out our colab.
For a quick usage of the model for generating new sequences by prompting it with a fasta file of homologous sequences check out ProtGPT3-MSA API.

Model Modalities

Aligned vs unaligned mode:ProtGPT3-MSA has been trained to process concatenated sets of homologs in both "aligned" (i.e., the homologs are passed aligned with gap tokens) and "unaligned" mode via special <gap> and <no_gap> tokens, which should be placed at the start of the concatenated protein sequences to select the modality.
N-to-C vs C-to-N:ProtGPT3-MSA has been trained to process concatenated homologs in both N-to-C and C-to-N directions, via two special "directional" tokens, "1" for N-to-C and "2" for C-to-N which which should be placed at the start of the concatenated protein sequences (i.e., before the gap token) to select the direction.

We provide some examples below.

Uses

How to Get Started with the Model

Install dependencies:

pip install transformers accelerate torch

Load the model and tokenizer:

python

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
import random
import re

# ---- Intialise useful methods to prompt ProtGPT3-MSA ----
def process_style(seq: str, gap: bool):
    """Remove gaps, uppercase insertions, drop X."""
    if gap:
        # keep gaps
        return re.sub(r"[X]", "", seq.upper())
    else:
        # remove gaps
        return re.sub(r"[X]", "", seq.replace("-", "").upper())

def build_prompt(
    sequences: list, 
    gap: bool = False,
    direction: str ="1"
) -> str:
    """Build prompt for ProtGPT3-MSA
    Args:
      sequences: list of up to 15 homologous protein sequences (i.e., each entry in the list should be a homolog)
      gap: if True, process sequence in the aligned mode, so the homologs in sequences should be aligned (i.e., same length with gap tokens) if False sequences can be unaligned.
      direction: direction in which sequences should be processed/generated, "1": N-to-C, pass "2" to generate homologs in reversed C-to-N direction, importantly sequences should also be reversed if direction="2"  
    """
    assert len(sequences) <= 15, "The model cannot be prompted with more than 15 sequences (i.e., reduce the number of sequences to 15 or less)"

    # randomise order of sequences
    random.shuffle(sequences)

    if gap:
        gap_token = "<gap>"
        assert all(len(s) == len(sequences[0]) for s in sequences), "Sequences in the prompt have different len(), but should be aligned, either align them or use no_gap mode"
    else:
        gap_token = "<no_gap>"

    tokens: List[str] = ["<|bos|>", direction, gap_token]
    for seq in sequences:
        # add separator token between sequences
        tokens.append("<s>")
        tokens.extend(list(process_style(seq,gap=gap)))

    # Match train-time separator before continuation
    tokens.append("<s>")
    return " ".join(tokens)
## --------------------------------------

model_id = "AI4PD/ProtGPT3-MSA"

# Load tokenizer for generation
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True,add_bos_token=False, add_eos_token=False, padding_side="left") # BOS token manually added in build_prompt

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

model.eval()

Few-shot generation with unaligned homologs

Use the <no_gap> modality token for unaligned sequences. Separate homologous sequences with the <s> separator token.

python

import torch


homologs = [
    "MKTAYIAKQRQISFVKSHFSRQDILD",
    "MKTVYIAKQRQISFVKSHFSRQDILD",
    "MKTAYIAKQRQINNVKSHFSRQNILD",
    # Add up to 15 homologous protein sequences
]

prompt = build_prompt(sequences=homologs)

inputs = tokenizer(prompt, return_tensors="pt", padding=True).to(model.device)

with torch.no_grad():
    output_ids = model.generate(
        inputs["input_ids"],
        max_new_tokens=512, # CHANGE to desire length (i.e., protein length times n. of generated homologs sequentially)
        do_sample=True,
        temperature=0.8,
        top_p=0.9,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.pad_token_id,
        num_return_sequences=20, # set to desired number of protein sequences to be generated in parallel
    )

generated = tokenizer.decode(output_ids[0], skip_special_tokens=True)
# split sequences generated sequentially
segments = generated.split("<s>")
# print each sequence
for s in segments:
  print(s.replace(" ",""),"\n")

Few-shot generation with aligned homologs

Use the <gap> modality token for aligned sequences. Gap characters may be included in the prompted sequences.

python

import torch

# must have the same length and be aligned
aligned_homologs = [
    "MKTAYIAKQRQI--SFVKSHFSRQDILD",
    "MKTVYIAKQRQI--SFVKSHFSRQDILD",
    "MKTAYIAKQRQINNSFVKSHFSRQNILD",
]

prompt = build_prompt(sequences=aligned_homologs, gap=True)

inputs = tokenizer(prompt, return_tensors="pt", padding=True).to(model.device)

with torch.no_grad():
    output_ids = model.generate(
        inputs["input_ids"],
        max_new_tokens=512, # CHANGE to desire length (i.e., protein length times n. of generated homologs sequentially)
        do_sample=True,
        temperature=0.8,
        top_p=0.9,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.eos_token_id,
        num_return_sequences=20, # set to desired number of protein sequences to be generated in parallel 
    )


generated = tokenizer.decode(output_ids[0], skip_special_tokens=True)
# split sequences generated sequentially
segments = generated.split("<s>")
# print each sequence
for s in segments:
  print(s.replace(" ",""),"\n")

Notes on prompting

Use <no_gap> for unaligned homologous sequences.
Use <gap> for aligned MSA-style inputs containing gap characters.
Separate protein sequences with <s>.
Provide up to 15 homologous sequences as context.
Sampling parameters such as temperature and top_p can affect sequence quality, diversity, and family consistency.
Generated sequences should be validated before experimental use.
Change max_new_tokens in generate() to control the number of protein generated sequentially (i.e., you are passing 15 homologs as prompt, this should roughly equal the length of a single protein).
- Use num_return_sequences in generate() to control the number of protein generated in parallel given the same prompt.

Out-of-Scope Use

The model should not be used as the sole basis for experimental, clinical, environmental, or safety-critical decisions. Generated sequences require downstream computational and experimental validation. The model is not guaranteed to generate functional, soluble, safe, synthesizable, or experimentally successful proteins.

The model should not be used for irresponsible or harmful biological design applications.

Bias, Risks, and Limitations

ProtGPT3-MSA learns from public protein sequence and MSA datasets and may reproduce biases present in those datasets. The model depends on the quality, relevance, and diversity of the homologous sequences provided in the prompt. Poor, unrelated, noisy, contaminated, or incorrectly aligned prompts may reduce generation quality.

Generated sequences may be nonfunctional, unstable, insoluble, repetitive, low-complexity, or biologically implausible. As with other generative protein models, ProtGPT3-MSA may present dual-use risks if applied irresponsibly.

Recommendations

Users should provide high-quality homologous protein sequences and validate generated sequences with appropriate downstream computational and experimental methods. For family-conditioned generation, users should carefully curate prompts and assess generated sequences using task-relevant criteria such as sequence identity, structural confidence, family-level consistency, solubility, and functional plausibility.

Training Details

Training Data

ProtGPT3-MSA was trained on approximately 8.5M MSAs from the OpenProteinSet Uniclust30 dataset. From each MSA, 16 sequences were sampled without replacement and concatenated in random order. This process was repeated 15 times for each MSA, resulting in approximately 560B training tokens.

Technical Specifications

Model Architecture and Objective

ProtGPT3-MSA is a decoder-only autoregressive protein language model using a Mixtral-style sparse Mixture-of-Experts architecture. It was trained to model concatenated sets of related protein sequences, enabling homolog-conditioned generation through prompting.

The model processes up to 16 concatenated protein sequences and supports both aligned and unaligned modalities. During inference, users may provide up to 15 homologous sequences and generate an additional sequence conditioned on the prompt.

Citation

BibTeX:

bibtex

@article{garibbo2026protgpt3,
  title={ProtGPT3: an Open-source family of Promptable and Aligned Protein Language Models},
  author={Garibbo, Michele and Boxo Corominas, Gerard and Stocco, Filippo and Illanes Vicioso, Ramiro and Middendorf, Lasse and Ferruz, Noelia},
  journal={bioRxiv},
  pages={2026--06},
  year={2026},
  publisher={Cold Spring Harbor Laboratory}
}

More Information

All models and code are released through the Hugging Face ecosystem and accompanying code repository.